From d76dc8d7c77bc09d88e6584cd6675a48b074adc6 Mon Sep 17 00:00:00 2001
From: DLANSAMA <dylanrayreed70@gmail.com>
Date: Wed, 26 Aug 2026 10:04:28 -0400
Subject: [PATCH 1/2] docs: fold Unreleased into 0.5.0

---
 CHANGELOG.md | 576 ++++-----------------------------------------------
 1 file changed, 46 insertions(+), 530 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 2f31df3..83dd4be 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,29 +5,41 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version
 
 ## [Unreleased]
 
-### Fixed
-
-- Windows `--json`: local path fields (`file`, `path`, `output`, `local_path`,
-  `workdir`, `config_path`, the `job` step paths, …) are emitted with `/`
-  separators instead of `\`. A single envelope previously mixed `\` in its
-  typed path fields with the `/` used by human-facing messages, remote printer
-  paths, and archive entries, so a consumer could not compare or join two path
-  fields without knowing which produced which. `~` home compaction is
-  unchanged. No effect on macOS/Linux.
+## [0.5.0] - 2026-08-26
 
-- User-facing docs now match shipped behaviour for `--confirm` on `job` /
-  `send` (upload still runs; exit `0` `uploaded_not_printed`), the fail-closed
-  camera streamer (opt-in, not auto-fallback), `doctor` fingerprint/`-v`
-  output, STEP/STP slice precedence, and sdist-relative links to repo-only
-  quality docs.
+### Added
 
+- **`plate tui` — a full-screen terminal UI** (optional extra:
+  `pip install 'platecli[tui]'`). A live dashboard (printer state, temperatures,
+  progress, AMS trays), a guided prepare screen (source validation, material and
+  quality presets with the AMS-loaded filament pre-selected, supports, slice +
+  time/filament preview), an explicit print-confirmation dialog (start / upload
+  only / cancel), and a live job monitor that follows a print to its terminal
+  state. It is a front-end over the existing pipeline — source validation, AMS
+  detection, download/slice, and the `job` request are the very same shared code
+  `plate go` runs, so the two cannot drift. Safety is unchanged: a print starts
+  only from the confirm dialog, upload-only leaves the file unstarted,
+  cancelling preserves the sliced file and says where it is, leaving the monitor
+  never stops a print, and quitting is refused while an upload is in flight.
+  `plate go` is unaffected and remains the no-extra-dependency path for SSH,
+  dumb terminals, and screen readers. Like `go`, `tui` is interactive-only:
+  `--json` and a non-TTY stdin exit `5` with the standard error envelope.
+  The prepare screen also offers **advanced slice settings** (`s`): a grouped
+  form over the named `slice` flags, plus an *Add an override* editor (a key, a
+  process/filament bucket, and a value) for everything the named flags do not
+  cover, recorded as `--set` / `--set-filament`. Blank fields keep profile
+  defaults, unsafe values are refused by the same validation the CLI applies,
+  and touching nothing leaves the slice byte identical to before. Use
+  `plate slice --list-settings` to see which keys your profiles accept.
 ### Removed
 
 - Python 3.9 support. The floor is 3.10 so every install resolves patched
   `zeroconf` (no 3.9-only advisory residual).
-
 ### Changed
 
+- 3MF estimate parsing walks the zip namelist in one pass and stops reading a
+  gcode header once both time and filament weight are known.
+
 - `upload` / `job` `--json` include `size_verified` (`false` when the printer
   omitted FTPS `SIZE` after the transfer). Mismatch is still a failure.
 
@@ -66,35 +78,6 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version
   The snapshot command moved out of `protocols/camera.py` into
   `bambu_cli.commands.snapshot`.
 
-## [0.5.0] - 2026-08-05
-
-### Added
-
-- **`plate tui` — a full-screen terminal UI** (optional extra:
-  `pip install 'platecli[tui]'`). A live dashboard (printer state, temperatures,
-  progress, AMS trays), a guided prepare screen (source validation, material and
-  quality presets with the AMS-loaded filament pre-selected, supports, slice +
-  time/filament preview), an explicit print-confirmation dialog (start / upload
-  only / cancel), and a live job monitor that follows a print to its terminal
-  state. It is a front-end over the existing pipeline — source validation, AMS
-  detection, download/slice, and the `job` request are the very same shared code
-  `plate go` runs, so the two cannot drift. Safety is unchanged: a print starts
-  only from the confirm dialog, upload-only leaves the file unstarted,
-  cancelling preserves the sliced file and says where it is, leaving the monitor
-  never stops a print, and quitting is refused while an upload is in flight.
-  `plate go` is unaffected and remains the no-extra-dependency path for SSH,
-  dumb terminals, and screen readers. Like `go`, `tui` is interactive-only:
-  `--json` and a non-TTY stdin exit `5` with the standard error envelope.
-  The prepare screen also offers **advanced slice settings** (`s`): a grouped
-  form over the named `slice` flags, plus an *Add an override* editor (a key, a
-  process/filament bucket, and a value) for everything the named flags do not
-  cover, recorded as `--set` / `--set-filament`. Blank fields keep profile
-  defaults, unsafe values are refused by the same validation the CLI applies,
-  and touching nothing leaves the slice byte identical to before. Use
-  `plate slice --list-settings` to see which keys your profiles accept.
-
-### Changed
-
 - **`plate tui` prepare screen is now two columns.** The form (source, material,
   quality, supports, the Prepare/Settings buttons) sits on the left and what the
   run produced — status, the time/filament estimate, and *Start print…* — on the
@@ -138,9 +121,22 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version
   `_underscore` helpers from the CLI entrypoint (roadmap B.4). Behaviour,
   output, JSON envelopes, and exit codes are unchanged; `cli.py` re-imports the
   same helpers from their new homes.
-
 ### Fixed
 
+- Windows `--json`: local path fields (`file`, `path`, `output`, `local_path`,
+  `workdir`, `config_path`, the `job` step paths, …) are emitted with `/`
+  separators instead of `\`. A single envelope previously mixed `\` in its
+  typed path fields with the `/` used by human-facing messages, remote printer
+  paths, and archive entries, so a consumer could not compare or join two path
+  fields without knowing which produced which. `~` home compaction is
+  unchanged. No effect on macOS/Linux.
+
+- User-facing docs now match shipped behaviour for `--confirm` on `job` /
+  `send` (upload still runs; exit `0` `uploaded_not_printed`), the fail-closed
+  camera streamer (opt-in, not auto-fallback), `doctor` fingerprint/`-v`
+  output, STEP/STP slice precedence, and sdist-relative links to repo-only
+  quality docs.
+
 - MQTT print-path integrity: `plate print` no longer reports a false
   `Print started` (exit 0) when the printer connection is refused. A refused
   CONNACK (e.g. wrong/rotated LAN access code) is delivered asynchronously via
@@ -297,7 +293,6 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version
   (`failed_step: "extract"`) instead of escaping as an unstructured error.
 - `plate job/send <model> --dry-run`: a 0-byte/unreadable sliceable model now fails
   the dry-run, matching the existing printer-ready empty-file check (dry-run parity).
-
 ### Documentation
 
 - Troubleshooting gained the three `plate tui` symptoms, keyed to the exact strings
@@ -321,9 +316,12 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version
 - Measured coverage/test numbers refreshed across the quality roadmap and test
   backlog from the current matrix (Linux 88.4%, Windows 88.09%, macOS 88.33%),
   replacing figures dating from before the TUI work.
-
 ### Security
 
+- `read_3mf_estimate` now refuses `Metadata/slice_info.config` members larger
+  than 10 MB and falls back to the gcode header, so a zip-bomb / oversized XML
+  in a downloaded `.3mf` cannot exhaust memory.
+
 - `insecure_tls` is now coerced strictly and fails **closed**. A hand-edited
   JSON string such as `"insecure_tls": "false"` (or `"no"`/`"0"`) is a truthy
   Python string and previously silently DISABLED TLS certificate validation for
@@ -371,486 +369,4 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version
 
 ## [0.4.0] - 2026-07-29
 
-### Added
-
-- Foundations for interactive guided-print mode (`plate go`).
-  This is internal groundwork only — no new subcommand, no prompts, and no change
-  to any existing command's behavior. The wizard itself follows in a later release.
-  - `bambu_cli/slicer/estimate.py` reads print time and filament weight from an
-    OrcaSlicer-produced `.3mf`, so a future preview can show an estimate before
-    anything is sent to the printer. It prefers `Metadata/slice_info.config` and
-    falls back to the header of `Metadata/plate_N.gcode`. It never raises —
-    callers get `Estimate(None, None)` on any failure — and time and weight
-    degrade independently, so a half-readable file still yields the half it has.
-    Implausible values (non-positive, over 30 days, over 10 kg) are treated as
-    unparsed on the principle that a wrong estimate is worse than a missing one.
-  - `bambu_cli/interactive/presets.py` maps friendly material and quality choices
-    onto the flags `job` already accepts, so the wizard can expose zero slicer
-    knobs. Material presets carry their own nozzle and bed temperatures, sourced
-    from the Bambu `@base` filament profiles.
-- Interactive guided-print wizard `plate go`. It walks you from a model URL (or
-  local file) to a running print without touching a slicer: paste a source, confirm
-  the printer, pick a material and quality preset, answer one supports question, see
-  a time/filament preview, then a default-No confirm before anything is sent. It is a
-  front-end over the existing pipeline — it drives the same `download` → `slice` →
-  `job` machinery `plate job` uses, so slicing is bit-identical. Notes:
-  - Interactive by design: `go` requires a TTY on stdin. Piped/non-interactive stdin
-    is refused with a pointer to `plate job <url> --confirm` for scripts, and `--json`
-    always emits the error envelope (schema `docs/schemas/go.json`) and exits 5 —
-    interactive mode has no machine contract. `go` is a local command, so an
-    unconfigured printer reaches the wizard, which offers to run `plate setup` itself
-    rather than hard-failing at the network gate.
-  - The confirm step defaults to No; answering Yes is the deliberate-action gate,
-    equivalent to `job --confirm`. Declining offers an upload-only path, and declining
-    that keeps the sliced file rather than deleting it.
-  - New package `bambu_cli/interactive/` (`prompts.py`, the only module that touches
-    interactive input, and `session.py`, the linear state machine). Both the prompt
-    layer and the pipeline collaborators are injectable, so the whole flow is tested
-    without a TTY, a printer, a network, or a real slicer.
-  - Bare `plate` (no subcommand) now launches the wizard when stdin **and** stdout
-    are both TTYs and `--json` was not passed. Every non-interactive context — CI,
-    pipes, `subprocess`, `plate | less`, `--json` — keeps the previous behavior
-    (help to stderr, exit `5`), so no script or agent can observe a change. The
-    top-level `--help` epilog now advertises `plate go`.
-  - The material step is AMS-aware: it reads the loaded filament from printer status
-    and, when it matches a preset (PLA/PETG/ABS/TPU), offers it as the prompt default
-    marked "(detected in AMS)". The read is best-effort through the existing status
-    machinery — any MQTT error, timeout, missing AMS, or unknown material falls back
-    silently to the plain PLA default and never blocks the wizard.
-- `camera_direct_only` config key (default `false`): when set, disables the Docker/RTSP streamer fallback so any direct port-6000 grab failure aborts instead of silently falling through to the unauthenticated streamer. Closes the remaining camera fallback residuals noted in SECURITY.md. X1-series printers still need the streamer; unset `camera_direct_only` for them.
-- Published JSON schemas for the last four `--json` commands that lacked them:
-  `upload.json`, `files.json`, `stop.json`, `setup.json`. README has claimed
-  "every command speaks `--json` with published schemas" for a while; that is now
-  true, and enforced rather than asserted.
-
-### Changed
-
-- `main` now carries a `.devN` version between releases (`0.4.0.dev0`). It kept the
-  released version after 0.3.0, so `plate --version` from a source checkout printed
-  `0.3.0` while running seven commits of unreleased code — and the bug-report
-  template asks users to paste exactly that string. `docs/releasing.md` gains a
-  post-release step so this cannot recur, and step 1 now says explicitly to drop the
-  suffix when releasing, since `release.yml` compares the tag to `pyproject.toml`
-  exactly.
-- Schema coverage is derived instead of hand-maintained. The contract suite reads
-  the subcommand list from `build_parser()` and fails if a command has no schema,
-  if a schema's `$id` disagrees with its filename, if a schema file is
-  unreferenced, or if the `docs/api.md` table omits one. The previous
-  hand-written list had already drifted — it omitted `status.json`.
-- `status_event.json` and `version.json` gained the `$id` and `title` fields the
-  other schemas already carried, so every published schema is self-identifying.
-- CI coverage floor raised **81 → 83**. Measured coverage had reached 84.10% on
-  Linux while the gate still accepted 81, leaving three points of silent
-  regression room; the existing drift guards pin the test count and the floor but
-  never the measured percentage. 83 rather than 84 because **Windows is the
-  binding leg at 83.85%** (Linux 84.10%, macOS 84.08%, Linux 3.9/3.12 ~84.3%), so
-  a floor of 84 would fail CI on Windows.
-
-### Fixed
-
-- Material presets for the upcoming `plate go` wizard name their filament profile
-  in full (`Bambu ABS @base`) rather than by bare material (`ABS`). `slice` matches
-  `--filament` as a case-insensitive substring against every `@base` profile and
-  takes the first `os.listdir` hit, so a bare `ABS` also matches
-  `Bambu Support for ABS @base.json` — support-interface filament — and `TPU`
-  matches `Generic TPU for AMS`. Because `os.listdir` order is filesystem-dependent,
-  the profile chosen could differ between Linux and Windows. Caught before the
-  wizard shipped; a regression test pins each preset to exactly one profile. The
-  underlying substring matching in `slice`/`job` is unchanged and still matches
-  loosely by design — only the preset values are now unambiguous.
-
-- `tests/privacy_smoke.py` no longer treats the tokens of a GitHub noreply email
-  as local account names. It split `user.email` on every non-alphanumeric
-  character, so a `<id>+<login>@users.noreply.github.com` address produced
-  `users`/`noreply`/`github` patterns that flagged every tracked file containing
-  the word "users" — destroying the "any tracked-file hit is a real leak" triage
-  rule. Email candidates are now mined only from the local part (before `@`),
-  numeric id prefixes are dropped, and `users`/`noreply`/`github` are in
-  `GENERIC_LOCAL_NAMES` as a backstop. Covered by a new hermetic regression test
-  (`tests/test_privacy_smoke_patterns.py`).
-- Auto-repair of URL-derived filenames is now correct in four ways it was not, and
-  the repairer is guaranteed to produce names the printer-side check accepts.
-  Previously `_sanitize_download_filename` could emit a name `_safe_remote_name`
-  refused, meaning a model could download and then fail to upload.
-  - **Windows device names are detected before the first dot.** The check used
-    `os.path.splitext`, which strips only the last extension, so `aux.gcode.3mf`,
-    `con.gcode.3mf` and friends passed both functions — and `.gcode.3mf` is the
-    print-ready format this tool produces. On Windows those names are unusable.
-  - **The length cap is enforced in UTF-8 bytes, not characters.** 160 CJK
-    characters is 480 bytes and exceeded ext4's 255-byte limit, so the repairer
-    produced names the local filesystem rejected with `ENAMETOOLONG`. Truncation
-    stops on a codepoint boundary, so no mojibake.
-  - **A pathological extension no longer defeats truncation.** `"x." + "a" * 200`
-    left the stem limit at 1 and returned all 202 bytes untruncated.
-  - **Trailing whitespace/dots are stripped after truncating, not before**, since
-    truncation can land on a space — which Windows drops and the safety check
-    rejects.
-  Repaired names differ from before for these inputs (e.g. `aux.gcode.3mf` now
-  becomes `_aux.gcode.3mf`), so a re-download may land under a new name. Ordinary
-  names such as `USB-C Cover.stl` are untouched, and `#`/`%` remain legal.
-- `slice --list-settings` writes its listing to **stdout** instead of stderr, one
-  key per line and unwrapped. It went through `logger.info`, so the obvious usage
-  — `plate slice --list-settings | grep layer_height` — silently matched nothing,
-  and Rich wrapped long values such as `description` across lines, defeating
-  `grep` even when stderr was captured. The header and closing hint stay on stderr
-  as human chrome, so redirecting stdout to a file yields only data. `--json`
-  output is unchanged.
-- `plate setup` no longer silently deletes config keys it does not manage.
-  It rebuilt `config.json` from scratch, so re-running it for an unrelated reason
-  dropped hand-added settings — `camera_port`, the `*_timeout` tunables, and
-  security opt-ins such as `camera_direct_only`, which looked enabled afterwards
-  but was not. Setup now preserves unmanaged keys and lists the ones it kept.
-  Keys it *does* manage stay authoritative, so switching an inline `access_code`
-  to an `access_code_file` still removes the plaintext secret, and declining
-  `insecure_tls` still turns it off.
-
-## [0.3.0] - 2026-07-26
-
-### Added
-- `tests/fixtures/cube.stl`, a 20 mm ASCII-STL cube — the only model committed
-  to the repo (`*.stl` / `*.3mf` remain gitignored, with a narrow negation for
-  `tests/fixtures/`). Gives `job` / `print` something to exercise by hand
-  without dropping a stray model into the tree.
-- AGENTS.md now states the stdout/stderr contract for `--json` (envelope on
-  stdout, Rich logs and `-v` diagnostics on stderr; parse stdout only) and warns
-  agents to confirm whether they are running an installed `plate` or the
-  checkout before reporting behaviour.
-- `docs/troubleshooting.md`: a symptom-keyed troubleshooting guide organised
-  around the errors the CLI emits (MQTT/FTPS connectivity, cert-pin mismatches,
-  OrcaSlicer and profile paths, camera/Docker, SSRF-blocked downloads, config
-  location). Linked from the README, the user guide, and the bug-report
-  template; ships in the sdist.
-- `tests/test_docs_links.py`: checks that cross-doc GitHub links resolve to real
-  repo paths and that every in-page anchor in the troubleshooting guide matches a
-  real heading.
-- README: a "How it compares" section describing adjacent cloud-free Bambu
-  projects and when to use them instead.
-- Printer error codes are now shown in hex as well as decimal (e.g. `Print failed
-  with error code 83935248 (hex 0x0500C010)`), and JSON error envelopes gain an
-  additive `printer_error_code_hex` field. Bambu documents these codes in hex.
-- `tests/test_docs_consistency.py`: guards against version drift in README /
-  `docs/api.md` / the bug-report template, and against `docs/manual.md` sections
-  missing from its Contents list.
-- `docs/manual.md` documents that Printables downloads are performed on the
-  user's behalf, subject to Printables' terms and each model's own licence.
-
-### Changed
-- The in-development version now carries a `.devN` suffix (`0.3.0.dev0`).
-  Previously `main` advertised the same version as the last PyPI release while
-  behaving differently — `plate --version` could not distinguish a source build
-  from the release, which made bug reports ambiguous. Drop the suffix when
-  tagging.
-- `docs/manual.md`: the OrcaSlicer section now covers per-OS installation, the
-  exact auto-detected default paths, and how to override them.
-- platecli now identifies itself honestly when talking to Printables:
-  requests carry `User-Agent: platecli/<version> (+https://github.com/DLANSAMA/platecli)`
-  instead of impersonating Chrome, and the forged `Origin`/`Referer` headers
-  are gone. Verified against the live Printables GraphQL API. Generic
-  user-supplied download URLs keep a browser-compatible User-Agent for CDN
-  compatibility, now with the honest `platecli/<version>` token appended.
-- Outbound HTTP is throttled to at most one request per second per host, and
-  `429`/`503` responses are retried while respecting `Retry-After` (clamped to
-  30 s, at most two retries).
-- README: stated the OrcaSlicer and Python 3.9+ prerequisites, qualified non-P1
-  printer support as best-effort, clarified that `plate snapshot` grabs P1/A1
-  cameras directly while X1-series needs the Docker streamer, and added
-  unattended-printing safety and maintenance-expectations sections.
-- Removed hardcoded version strings from README, `docs/api.md`, and the bug-report
-  template; the version is single-sourced from `pyproject.toml`.
-- Refreshed measured test/coverage numbers in `docs/quality-roadmap.md` and
-  `docs/test-backlog.md` (758 tests, 84.03%).
-- `AGENTS.md` retitled to the public product name.
-- `docs/api.md`: the `pause.json` / `resume.json` bullets now describe both the
-  success and the `confirmation_required` shape, matching the dual schemas.
-- CI runs weekly on a schedule (external drift: firmware, Printables API,
-  OrcaSlicer, new CVEs) and pins ruff/mypy/bandit/pip-audit versions.
-- Dependabot config added for GitHub Actions and uv dependencies.
-- New `docs/releasing.md` with the yank/rollback runbook.
-
-### Changed (BREAKING)
-- `pause` and `resume` now require `--confirm`, matching `stop`/`print`/`gcode`/`delete`.
-  Pausing mid-print parks a hot nozzle over the part; resuming an abandoned print
-  restarts motion unsupervised. Without `--confirm` they now refuse with exit code `5`
-  and (with `--json`) `"status": "confirmation_required"`. Scripts calling
-  `plate pause` / `plate resume` must add `--confirm`.
-- `print` without `--confirm` now exits `5` (`EXIT_COMMAND_ERROR`) instead of `0`.
-  The JSON payload is unchanged (`"status": "confirmation_required"`). This aligns it
-  with `stop`/`gcode`/`delete` and with the exit-code table in `docs/api.md`; a script
-  that treated exit `0` as "print started" was already silently wrong.
-
-### Fixed
-- **`scripts/bambu.py` — the documented "run from a source checkout without
-  installing" entrypoint — never worked.** Python puts the script's own
-  directory (`scripts/`) on `sys.path`, not the repo root, so it died with
-  `ModuleNotFoundError: No module named 'bambu_cli'` on any clean checkout. It
-  now prepends the repo root before importing. This mattered more than it looks:
-  with the shim broken, the only way to run the CLI was an installed `plate`,
-  which is usually an older release than the checkout — so contributors and
-  agents were reporting the *released* behaviour as if it were the source's.
-- **`scripts/clean_artifacts.py` no longer deletes the developer's `.venv`.**
-  The script appended `.venv` to its removal list whenever `GITHUB_ACTIONS` was
-  unset — i.e. on every local run — and removed it with `ignore_errors=True`. On
-  Windows any locked file (a running `python.exe`) left a *partially* deleted
-  virtualenv, after which every `uv run` failed with `No pyvenv.cfg file` and
-  exit code 106, recoverable only by hand. Nothing warned about this, and the
-  name "clean_artifacts" does not suggest destroying the dev environment. The
-  deletion also served no purpose: `tests/release_readiness_smoke.py` skips any
-  path containing `.venv`, so a present `.venv` was never flagged. `.venv` is now
-  opt-in via `--venv` / `--all`, confirmed interactively (`-y` to skip), and the
-  walk no longer descends into `.git`, `.venv`, `venv`, `.claude` or
-  `node_modules`. Removal is verified instead of best-effort: a surviving path is
-  reported with recovery steps and a non-zero exit. Adds `--dry-run`, a docstring,
-  `argparse --help`, and `tests/test_clean_artifacts.py`. CI's invocation is
-  unchanged and never passed `--venv`.
-- **`job` now validates print options on every path, not just under `--confirm`.**
-  `_run_job` only checked `--use-ams` / `--ams-mapping` when `--confirm` was passed
-  and `--upload-only` was not, so the two paths agents use as a pre-check both
-  reported success for flags that would later be rejected. `job --dry-run
-  --ams-mapping 0,1` returned `dry_run_local_skipped` even though the same flags
-  fail without `--dry-run` — inconsistent with `--infill`, which was already caught
-  in dry-run. Worse, `job --upload-only --ams-mapping 0,1` uploaded and then handed
-  back a `next_command` of `["print", …, "--ams-mapping", "0,1"]` that `print`
-  rejects with exit `5`, so an agent following the documented handoff hit a
-  guaranteed failure *after* a real upload. Validation now runs for any supplied
-  print option and still passes when none are given.
-- **`status` no longer returns a partial MQTT delta as if it were a full state
-  snapshot.** The printer publishes incremental updates on the `report` topic and
-  only sends the complete state in reply to `pushall`, but `status` returned
-  whichever message arrived first. Mid-print that was intermittently a single
-  `nozzle_temper` reading, so `plate --json status` emitted a `printer` object
-  with no `gcode_state`, `mc_percent`, or `layer_num` — a random `KeyError` for
-  any agent reading those fields. Report messages are now merged into one
-  accumulated state and `status` keeps waiting (re-issuing `pushall` on each
-  retry, honouring the existing timeout and retry count) until the state is
-  complete. If only deltas ever arrive, it now fails with a clear error naming
-  the missing keys and exit code `6` instead of emitting an incomplete object.
-  `doctor` and `print --dry-run`, which use the reply only as a liveness probe,
-  still accept the first message.
-- `status --monitor` merges deltas the same way, so a temperature-only update no
-  longer streams as `"gcode_state": "UNKNOWN"` at 0% mid-print.
-- **`setup` and `--migrate-access-code` no longer fail outright when the config
-  directory is behind filesystem virtualization.** Windows MSIX/AppContainer
-  redirection of `%APPDATA%` makes `os.replace` report `ERROR_NOT_SAME_DEVICE`
-  (`WinError 17`) even for two paths in the *same* directory, so every config and
-  access-code write raised `Could not migrate access code: [WinError 17] ...` and
-  platecli could not persist configuration at all on such a machine. The atomic
-  temp-file-plus-rename path is unchanged and still the default; only on that
-  specific errno (`EXDEV` / winerror 17) does it fall back to writing the file in
-  place, and it warns that the write is not crash-safe. The fallback writes the
-  target directly rather than copying the temp file over it, so a secret is never
-  left in a second location. Reproduced and verified against a real redirected
-  `%APPDATA%\bambu`.
-- **OrcaSlicer auto-detection now finds stock Windows installs.** The current
-  Windows installer lays the binary down as `orca-slicer.exe`, but detection only
-  probed `OrcaSlicer.exe`, so `plate setup` missed every default install and users
-  had to set `orca_slicer` by hand. Both names are now probed under each install
-  root (hyphenated first), plus a `PATH` lookup for custom locations. Found on a
-  real Windows box with OrcaSlicer 2.4.2.
-- When no OrcaSlicer exists anywhere on the machine, the `orca-slicer` /
-  `profiles-dir` errors from `preflight`, `config validate`, and `slice` now append
-  a ready-to-run install command for the host platform (`winget` / `brew` /
-  `flatpak`) instead of only suggesting a config edit — there is nothing to point
-  the config at until something is installed. When an install *is* present, the
-  existing "detected at <path>" hint still wins.
-- `tests/live_printer_smoke.py` now decodes CLI subprocess output as UTF-8 instead of
-  the platform default. On a Windows box with a cp1252 codepage the harness's reader
-  threads died with `UnicodeDecodeError` on the CLI's emoji output, losing stdout/stderr
-  while the run still reported success. `tests/agent_cli_smoke.py` already carried this
-  fix; the live harness had been missed.
-- README, AGENTS.md, SECURITY.md, docs/api.md, and docs/manual.md no longer claim a
-  universal `--confirm` gate that `pause`/`resume` did not implement; `--confirm` is
-  now documented as a deliberate-action gate rather than an authorization boundary.
-- `docs/schemas/pause.json` / `resume.json` accept the `confirmation_required` status.
-- Filenames containing square brackets (e.g. `part[v2].stl`, common in Printables model titles) no longer crash the CLI or get silently truncated in log output — the Rich log handler no longer parses log text as markup.
-- `--json` now always writes a parseable error envelope to stdout: the JSON envelope is emitted before the human-readable log line, so a logging failure can no longer leave stdout empty.
-- The envelope-before-log ordering now covers the whole package (`commands/`,
-  `camera`, `download/`, `slicer/`, `setup_cmd/`), not just `cli.py`. Every log call on
-  an error path now also routes through the shared `logging_utils.safe_log_error`, which
-  degrades to a bare stderr write instead of propagating, so a handler that raises while
-  rendering a user-controlled string can no longer swallow the `--json` envelope or
-  replace the real `BambuError` with a logging traceback.
-- `plate doctor`'s reported `camera_snapshot_note` (and `camera.py`'s snapshot
-  docstring) claimed P1P/P1S snapshots go through the BambuP1Streamer container.
-  It is the reverse: P1/A1-class cameras are captured directly with no Docker,
-  and the container is the X1-series fallback.
-- `plate preflight` now says what to do when `orca_slicer` / `profiles_dir` are not
-  configured, instead of printing `OrcaSlicer not found at` with a blank path.
-- Contributor setup uses `uv sync --extra test`, so the documented test command works
-  on a fresh clone (`uv sync` alone does not install pytest).
-- Config and access-code writes are now atomic (temp file + fsync + rename), so a crash
-  mid-write can no longer truncate `config.json` or the access-code file. `config.json`
-  also keeps a `.bak` of the previous file; access-code files are written atomically at
-  0600 but without a backup copy (fewer copies of credentials on disk). Mode stays `0600`.
-- `plate slice` now says what to do when `orca_slicer` is not configured at all, instead
-  of reporting `OrcaSlicer not found at ` with a blank path.
-- Device, file, print and G-code commands now emit the `--json` error envelope before
-  writing the human-readable log line, so a failure in the logging layer can no longer
-  leave stdout without a parseable envelope.
-- **A UTF-8 byte-order mark on `config.json` silently voided the entire config.**
-  PowerShell's `Set-Content -Encoding utf8`, Notepad's "Save as UTF-8", and most
-  Windows editors write a BOM. `json.load` then failed, the error was swallowed,
-  and every command reported `Printer IP is not configured. Please run 'plate
-  setup' first.` — advice that would have overwritten the config the user was
-  trying to repair. Config files are now read as `utf-8-sig`, which handles BOM
-  and BOM-less files identically.
-- A config that exists but fails to parse is no longer reported as *missing*.
-  `load_config` always logs the real parse error, and `plate preflight` now says
-  the file exists and could not be read rather than `Config not found at …`.
-- `plate preflight`'s Docker check is now model-aware: it no longer warns P1/A1
-  owners that "camera snapshots will be unavailable" when those printers capture
-  directly over the printer's TLS camera port. (The matching `doctor` note was
-  already corrected; this was the remaining copy of the same wrong claim.)
-- `bambu_cli/bambu.py` looks up `reconfigure` via `getattr`, fixing the sole
-  outstanding mypy error (`TextIO` has no `reconfigure`) without changing the
-  Windows encoding hardening it performs.
-- `.gitignore` now ignores `result.json`, the slice-result file OrcaSlicer's CLI
-  drops in the working directory on every `plate slice`. The existing entry read
-  `results.json` (plural) and never matched, so the file showed up as untracked
-  in every working tree after a slice.
-
-### Security
-- **`plate doctor` no longer prints your printer's LAN IP or full certificate
-  SHA-256 in its human output by default.** doctor output is routinely pasted
-  into issue reports and recorded into the README/PyPI demo GIFs. The IP now
-  reads `<redacted>`, and once a cert is pinned doctor prints a hex-free match
-  confirmation instead of the fingerprint. Pass `-v`/`--verbose` to see both.
-  The full fingerprint is still printed when the cert is *not* yet pinned (you
-  need it to pin), the MQTT-failure message still echoes the configured IP so a
-  typo is diagnosable, and the `--json` contract is unchanged.
-- The bug-report template now asks for `plate doctor --json` (which redacts the
-  LAN IP) instead of the human output.
-- All GitHub Actions are pinned to immutable commit SHAs, enforced by
-  `tests/ci_workflow_smoke.py`. Notably `pypa/gh-action-pypi-publish` was
-  tracking the mutable `release/v1` branch in the job that holds the PyPI
-  trusted-publishing `id-token: write` permission.
-- Releases now re-run the full CI matrix on the tagged commit before publishing,
-  so a tag on a red commit can no longer reach PyPI.
-- The four demo GIFs committed to the repo (`docs/demo-{dark,light}.gif`,
-  `docs/doctor-{dark,light}.gif`) previously had the maintainer's printer LAN IP
-  and full certificate SHA-256 **rasterised into the frames**, where no
-  code-level redaction could reach them. The GIFs are embedded in README.md and
-  docs/manual.md and served from raw.githubusercontent.com, making them the most
-  widely-read copy of that data. Re-recorded against the doctor redaction from
-  #61; the fingerprint now shows a hex-free match confirmation and the IP reads
-  `<redacted>`.
-
-## [0.2.2] - 2026-07-24
-
-### Added
-- `plate snapshot --unique`: timestamped output filenames
-  (`printer_snapshot_20260724T195820Z.jpg`) so repeated captures never
-  overwrite each other; composes with `--output` by inserting the stamp
-  before the extension. Designed for AI-agent workflows.
-- Snapshot `--json` output now includes `captured_at` (ISO-8601 UTC) and
-  `sha256` (digest of the JPEG bytes) so callers can verify a capture is
-  genuinely new. Additive fields; `docs/schemas/snapshot.json` updated.
-- Camera-snapshot guidance for agents in `AGENTS.md` and a Camera snapshots
-  section in `docs/manual.md`.
-
-## [0.2.1] - 2026-07-24
-
-### Fixed
-- **Slice temperature safety now covers process-section overrides**: nozzle/bed
-  temperature keys passed via `--set` or the `"process"` object of
-  `--settings-json` are validated against the same printer-safety bounds as
-  filament overrides (previously only the filament path was checked).
-- `plate doctor` honors `--network-timeout` for its MQTT, FTPS, and version
-  probes (previously hardcoded 5 s with silent retries taking ~23 s), and no
-  longer retries the MQTT check.
-- Deprecation and setup messages referenced the pre-rename `bambu setup`
-  command; they now correctly say `plate setup`.
-- The headless (non-TTY) setup error now includes a copy-pasteable
-  non-interactive example command.
-
-### Added
-- `docs/schemas/status.json`: published JSON Schema for static
-  `plate status --json` output (the monitor stream already had
-  `status_event.json`), validated in contract tests.
-- Community files: code of conduct, issue forms (bug report asks for
-  `plate doctor` output), and a pull-request template.
-
-### Changed
-- README restructured into a concise landing page; full technical
-  documentation moved to `docs/manual.md` (ships in the sdist, linked from
-  the PyPI sidebar via the new `Documentation` project URL).
-- Quickstart examples use a real Printables model (3DBenchy) instead of a
-  placeholder URL.
-- Repo hygiene: `pytest.ini` and `.coveragerc` folded into `pyproject.toml` (`[tool.pytest.ini_options]`, `[tool.coverage.*]`); removed the `.jules/` bot-notes directory. Test/coverage behavior unchanged; sdists still carry the full test config via `pyproject.toml`.
-
-## [0.2.0] - 2026-07-23
-
-### Changed
-- **Project renamed to `platecli`; the installed command is now `plate`** (was
-  `bambu-cli`, published as `bambu-local-cli`). The rename removes the vendor
-  name from the project branding; platecli remains an unofficial tool for
-  Bambu Lab printers and is not affiliated with Bambu Lab. The internal Python
-  package (`bambu_cli`), environment variables (`BAMBU_*`), and config path
-  (`~/.config/bambu/`) are unchanged, so existing configs keep working —
-  only the command you type changes. The old PyPI release `bambu-local-cli`
-  0.1.0 is yanked to point users at `platecli`.
-
-## [0.1.0] - 2026-07-18
-
-Initial development version: LAN-mode printer control (MQTT/FTPS), one-shot `job`/`send` orchestration, OrcaSlicer integration, guided setup with mDNS discovery, camera snapshots, SSRF-safe downloads with Printables support, simulation mode, and agent-facing `--json` output.
-
-### Documentation
-- Full doc truth pass (2026-07-17 audit): aligned **AGENTS.md**, **CONTRIBUTING.md**,
-  **SECURITY.md**, **README.md**, **docs/api.md**, **docs/quality-roadmap.md**,
-  **docs/test-backlog.md**, live/mutation docs. Corrected stale claims (mypy residual
-  blocklist, ≥98% coverage). Expanded config reference in README; security known
-  limitations (camera Docker bind, pin soft-fallback, HTTP downloads, pause/resume).
-  Architecture grade noted as **A−** until domain→`cli` helper extraction lands.
-  Honest metrics: ~618 non-live tests, ~82% coverage, CI floor 81, target 92 for A+/1.0.
-- **Packaging doc policy:** PyPI sdist ships only user/agent docs (`README`, `AGENTS`,
-  `SECURITY`, `CHANGELOG`, `docs/api.md`, `docs/schemas/*`). Contributor planning docs
-  (`CONTRIBUTING`, quality-roadmap, test-backlog, mutation/live-smoke) stay GitHub-only;
-  enforced by `MANIFEST.in` + package_contents forbidden list. Wheel remains runtime-only.
-
-### Security
-- **Camera streamer now binds loopback by default.** `camera_port` defaults to `127.0.0.1:1985:1984` (was `1985:1984`), so the BambuP1Streamer container no longer publishes the unauthenticated printer camera feed on all host interfaces (`0.0.0.0`). Set `camera_port` to `0.0.0.0:1985:1984` to deliberately restore LAN access. An **already-running** container keeps its old binding until recreated — run `docker rm -f bambu_camera` (the CLI now warns when it detects a running container still bound to a non-loopback interface). `camera_port` is also validated, and the localhost stream URL is now derived correctly from host-qualified specs (`[HOST:]HOSTPORT:CONTAINERPORT`).
-
-### Fixed
-- **Camera pin fail-open:** a `cert_fingerprint` **mismatch** during the direct P1/A1 camera grab now hard-aborts `snapshot` (exit 2) instead of being swallowed by the broad fallback handler and silently retried through the Docker streamer, which connects without honoring the pin. A *missing* pin still legitimately falls through to the streamer (X1 path).
-- **Camera socket fd leak:** the direct-grab TLS socket is now closed via the wrapped `SSLSocket` (which owns the fd after `wrap_socket` detaches it) rather than the detached raw socket, so a successful snapshot no longer leaks an fd / emits a `ResourceWarning`.
-- Global `--json` placed *before* the subcommand is now honored by `status`, `light`, `pause`, and `resume` (they re-declared `--json` with an implicit `False` default that clobbered the global flag, so `bambu-cli --json status` silently emitted nothing).
-- `--allow-private-ips` now actually enables private/LAN downloads for that invocation (it was parsed but never applied to runtime settings).
-- `load_access_code` and domain handlers raise structured `BambuError` / `abort` instead of calling `sys.exit` (process exit is CLI entry only).
-
-### Changed
-- Release workflow now creates the GitHub Release only after the PyPI publish succeeds (build → publish → release job chain).
-- Sdists include `pytest.ini` so the shipped test suite runs with the project's marker/coverage config.
-- Simulation mode (`--sim status`) reports representative bed/nozzle targets, fan speed, and WiFi signal instead of `?` placeholders.
-- The missing-config error now names the exact command: ``Please run `bambu-cli setup` first.``
-- `VERSION` is resolved from package metadata / `pyproject.toml` only (no duplicate string in `constants.py`).
-- MQTT status-monitor teardown no longer uses a bare `except:`.
-- Removed `@mockable` / test-awareness indirection from production code.
-- CI coverage floor enforced at **81%** (`--cov-fail-under=81`, raised from 79; multi-OS minimum — Linux ~82.3% / Windows ~81.9% branch total); single pytest path; blocking purity greps for `sys.exit` / `@mockable` / Mock branches. (Further ratchet toward the 92% A+ target remains Phase C work in `docs/quality-roadmap.md`.)
-- Package renamed to `bambu-local-cli` for PyPI publication (the `bambu-cli` name on PyPI belongs to an unrelated project). The installed command remains `bambu-cli`.
-- Wheels no longer bundle non-runtime files (`README.md`, `AGENTS.md` inside the package).
-
-### Added
-- **Full-surface slicer overrides.** `slice` (and `job`/`send`) accept `--set KEY=VALUE` / `--set-filament KEY=VALUE` (repeatable) and `--settings-json '{"process":{…},"filament":{…}}'` to override **any** of the 176 OrcaSlicer process/filament settings, not just the ~17 with dedicated flags. Unknown keys warn (with a "did you mean" suggestion) but still pass through. Temperature overrides are re-validated against the printer-safety bounds so `--set-filament nozzle_temperature=999` is refused.
-- **Slicer setting discovery:** `slice --list-settings [--json]` dumps every settable process/filament key with an example value — the agent-facing way to learn the override vocabulary.
-- Named slicer convenience flags for the common tuning knobs (sugar over the generic override machinery), each verified against a real slice: `--layer-height`, `--first-layer-height`, `--brim`, `--speed`, `--seam-position {nearest,aligned,back,random}`, `--ironing {none,top,topmost,solid}`, `--support-threshold`, `--fan-speed`, and `--flow-ratio`.
-- JSON schemas under `docs/schemas/` and contract tests in `tests/contracts/`.
-- `docs/quality-roadmap.md` scoreboard and residual coverage policy.
-- TLS pin suite (`tests/test_tls_pinning.py`); SSRF / redirect residual coverage; security + contract pytest markers.
-- PyPI trusted publishing on tagged releases (`v*`).
-- `CHANGELOG.md`, `CONTRIBUTING.md`, `SECURITY.md`, Changelog project URL.
-
-### Tests
-- CLI e2e coverage for `--allow-private-ips` wiring into `RuntimeContext` / netsafety.
-- MQTT + FTPS certificate fingerprint pin match/mismatch (and deferred-handshake) suite.
-- Expanded unit coverage for netsafety handlers, setup helpers, slicer pure paths, wizard guided flows.
-
-### Removed
-- `requirements.txt`, which duplicated the `dependencies` already declared in `pyproject.toml`. Install with `uv pip install -e .` (or `pip install platecli`) instead.
-
-[Unreleased]: https://github.com/DLANSAMA/platecli/compare/v0.5.0...HEAD
-[0.5.0]: https://github.com/DLANSAMA/platecli/compare/v0.4.0...v0.5.0
-[0.4.0]: https://github.com/DLANSAMA/platecli/compare/v0.3.0...v0.4.0
-[0.3.0]: https://github.com/DLANSAMA/platecli/compare/v0.2.2...v0.3.0
-[0.2.2]: https://github.com/DLANSAMA/platecli/compare/v0.2.1...v0.2.2
-[0.2.1]: https://github.com/DLANSAMA/platecli/compare/v0.2.0...v0.2.1
-[0.2.0]: https://github.com/DLANSAMA/platecli/compare/v0.1.0...v0.2.0
-[0.1.0]: https://github.com/DLANSAMA/platecli/releases/tag/v0.1.0
+PLACEHOLDER_REST
\ No newline at end of file

From a9e868b9aaebd52e96c26f223a7a9104a9fcec40 Mon Sep 17 00:00:00 2001
From: DLANSAMA <dylanrayreed70@gmail.com>
Date: Wed, 26 Aug 2026 10:06:59 -0400
Subject: [PATCH 2/2] docs: fold Unreleased into 0.5.0

---
 CHANGELOG.md | 484 ++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 483 insertions(+), 1 deletion(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 83dd4be..d86e469 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -369,4 +369,486 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version
 
 ## [0.4.0] - 2026-07-29
 
-PLACEHOLDER_REST
\ No newline at end of file
+### Added
+
+- Foundations for interactive guided-print mode (`plate go`).
+  This is internal groundwork only — no new subcommand, no prompts, and no change
+  to any existing command's behavior. The wizard itself follows in a later release.
+  - `bambu_cli/slicer/estimate.py` reads print time and filament weight from an
+    OrcaSlicer-produced `.3mf`, so a future preview can show an estimate before
+    anything is sent to the printer. It prefers `Metadata/slice_info.config` and
+    falls back to the header of `Metadata/plate_N.gcode`. It never raises —
+    callers get `Estimate(None, None)` on any failure — and time and weight
+    degrade independently, so a half-readable file still yields the half it has.
+    Implausible values (non-positive, over 30 days, over 10 kg) are treated as
+    unparsed on the principle that a wrong estimate is worse than a missing one.
+  - `bambu_cli/interactive/presets.py` maps friendly material and quality choices
+    onto the flags `job` already accepts, so the wizard can expose zero slicer
+    knobs. Material presets carry their own nozzle and bed temperatures, sourced
+    from the Bambu `@base` filament profiles.
+- Interactive guided-print wizard `plate go`. It walks you from a model URL (or
+  local file) to a running print without touching a slicer: paste a source, confirm
+  the printer, pick a material and quality preset, answer one supports question, see
+  a time/filament preview, then a default-No confirm before anything is sent. It is a
+  front-end over the existing pipeline — it drives the same `download` → `slice` →
+  `job` machinery `plate job` uses, so slicing is bit-identical. Notes:
+  - Interactive by design: `go` requires a TTY on stdin. Piped/non-interactive stdin
+    is refused with a pointer to `plate job <url> --confirm` for scripts, and `--json`
+    always emits the error envelope (schema `docs/schemas/go.json`) and exits 5 —
+    interactive mode has no machine contract. `go` is a local command, so an
+    unconfigured printer reaches the wizard, which offers to run `plate setup` itself
+    rather than hard-failing at the network gate.
+  - The confirm step defaults to No; answering Yes is the deliberate-action gate,
+    equivalent to `job --confirm`. Declining offers an upload-only path, and declining
+    that keeps the sliced file rather than deleting it.
+  - New package `bambu_cli/interactive/` (`prompts.py`, the only module that touches
+    interactive input, and `session.py`, the linear state machine). Both the prompt
+    layer and the pipeline collaborators are injectable, so the whole flow is tested
+    without a TTY, a printer, a network, or a real slicer.
+  - Bare `plate` (no subcommand) now launches the wizard when stdin **and** stdout
+    are both TTYs and `--json` was not passed. Every non-interactive context — CI,
+    pipes, `subprocess`, `plate | less`, `--json` — keeps the previous behavior
+    (help to stderr, exit `5`), so no script or agent can observe a change. The
+    top-level `--help` epilog now advertises `plate go`.
+  - The material step is AMS-aware: it reads the loaded filament from printer status
+    and, when it matches a preset (PLA/PETG/ABS/TPU), offers it as the prompt default
+    marked "(detected in AMS)". The read is best-effort through the existing status
+    machinery — any MQTT error, timeout, missing AMS, or unknown material falls back
+    silently to the plain PLA default and never blocks the wizard.
+- `camera_direct_only` config key (default `false`): when set, disables the Docker/RTSP streamer fallback so any direct port-6000 grab failure aborts instead of silently falling through to the unauthenticated streamer. Closes the remaining camera fallback residuals noted in SECURITY.md. X1-series printers still need the streamer; unset `camera_direct_only` for them.
+- Published JSON schemas for the last four `--json` commands that lacked them:
+  `upload.json`, `files.json`, `stop.json`, `setup.json`. README has claimed
+  "every command speaks `--json` with published schemas" for a while; that is now
+  true, and enforced rather than asserted.
+
+### Changed
+
+- `main` now carries a `.devN` version between releases (`0.4.0.dev0`). It kept the
+  released version after 0.3.0, so `plate --version` from a source checkout printed
+  `0.3.0` while running seven commits of unreleased code — and the bug-report
+  template asks users to paste exactly that string. `docs/releasing.md` gains a
+  post-release step so this cannot recur, and step 1 now says explicitly to drop the
+  suffix when releasing, since `release.yml` compares the tag to `pyproject.toml`
+  exactly.
+- Schema coverage is derived instead of hand-maintained. The contract suite reads
+  the subcommand list from `build_parser()` and fails if a command has no schema,
+  if a schema's `$id` disagrees with its filename, if a schema file is
+  unreferenced, or if the `docs/api.md` table omits one. The previous
+  hand-written list had already drifted — it omitted `status.json`.
+- `status_event.json` and `version.json` gained the `$id` and `title` fields the
+  other schemas already carried, so every published schema is self-identifying.
+- CI coverage floor raised **81 → 83**. Measured coverage had reached 84.10% on
+  Linux while the gate still accepted 81, leaving three points of silent
+  regression room; the existing drift guards pin the test count and the floor but
+  never the measured percentage. 83 rather than 84 because **Windows is the
+  binding leg at 83.85%** (Linux 84.10%, macOS 84.08%, Linux 3.9/3.12 ~84.3%), so
+  a floor of 84 would fail CI on Windows.
+
+### Fixed
+
+- Material presets for the upcoming `plate go` wizard name their filament profile
+  in full (`Bambu ABS @base`) rather than by bare material (`ABS`). `slice` matches
+  `--filament` as a case-insensitive substring against every `@base` profile and
+  takes the first `os.listdir` hit, so a bare `ABS` also matches
+  `Bambu Support for ABS @base.json` — support-interface filament — and `TPU`
+  matches `Generic TPU for AMS`. Because `os.listdir` order is filesystem-dependent,
+  the profile chosen could differ between Linux and Windows. Caught before the
+  wizard shipped; a regression test pins each preset to exactly one profile. The
+  underlying substring matching in `slice`/`job` is unchanged and still matches
+  loosely by design — only the preset values are now unambiguous.
+
+- `tests/privacy_smoke.py` no longer treats the tokens of a GitHub noreply email
+  as local account names. It split `user.email` on every non-alphanumeric
+  character, so a `<id>+<login>@users.noreply.github.com` address produced
+  `users`/`noreply`/`github` patterns that flagged every tracked file containing
+  the word "users" — destroying the "any tracked-file hit is a real leak" triage
+  rule. Email candidates are now mined only from the local part (before `@`),
+  numeric id prefixes are dropped, and `users`/`noreply`/`github` are in
+  `GENERIC_LOCAL_NAMES` as a backstop. Covered by a new hermetic regression test
+  (`tests/test_privacy_smoke_patterns.py`).
+- Auto-repair of URL-derived filenames is now correct in four ways it was not, and
+  the repairer is guaranteed to produce names the printer-side check accepts.
+  Previously `_sanitize_download_filename` could emit a name `_safe_remote_name`
+  refused, meaning a model could download and then fail to upload.
+  - **Windows device names are detected before the first dot.** The check used
+    `os.path.splitext`, which strips only the last extension, so `aux.gcode.3mf`,
+    `con.gcode.3mf` and friends passed both functions — and `.gcode.3mf` is the
+    print-ready format this tool produces. On Windows those names are unusable.
+  - **The length cap is enforced in UTF-8 bytes, not characters.** 160 CJK
+    characters is 480 bytes and exceeded ext4's 255-byte limit, so the repairer
+    produced names the local filesystem rejected with `ENAMETOOLONG`. Truncation
+    stops on a codepoint boundary, so no mojibake.
+  - **A pathological extension no longer defeats truncation.** `"x." + "a" * 200`
+    left the stem limit at 1 and returned all 202 bytes untruncated.
+  - **Trailing whitespace/dots are stripped after truncating, not before**, since
+    truncation can land on a space — which Windows drops and the safety check
+    rejects.
+  Repaired names differ from before for these inputs (e.g. `aux.gcode.3mf` now
+  becomes `_aux.gcode.3mf`), so a re-download may land under a new name. Ordinary
+  names such as `USB-C Cover.stl` are untouched, and `#`/`%` remain legal.
+- `slice --list-settings` writes its listing to **stdout** instead of stderr, one
+  key per line and unwrapped. It went through `logger.info`, so the obvious usage
+  — `plate slice --list-settings | grep layer_height` — silently matched nothing,
+  and Rich wrapped long values such as `description` across lines, defeating
+  `grep` even when stderr was captured. The header and closing hint stay on stderr
+  as human chrome, so redirecting stdout to a file yields only data. `--json`
+  output is unchanged.
+- `plate setup` no longer silently deletes config keys it does not manage.
+  It rebuilt `config.json` from scratch, so re-running it for an unrelated reason
+  dropped hand-added settings — `camera_port`, the `*_timeout` tunables, and
+  security opt-ins such as `camera_direct_only`, which looked enabled afterwards
+  but was not. Setup now preserves unmanaged keys and lists the ones it kept.
+  Keys it *does* manage stay authoritative, so switching an inline `access_code`
+  to an `access_code_file` still removes the plaintext secret, and declining
+  `insecure_tls` still turns it off.
+
+## [0.3.0] - 2026-07-26
+
+### Added
+- `tests/fixtures/cube.stl`, a 20 mm ASCII-STL cube — the only model committed
+  to the repo (`*.stl` / `*.3mf` remain gitignored, with a narrow negation for
+  `tests/fixtures/`). Gives `job` / `print` something to exercise by hand
+  without dropping a stray model into the tree.
+- AGENTS.md now states the stdout/stderr contract for `--json` (envelope on
+  stdout, Rich logs and `-v` diagnostics on stderr; parse stdout only) and warns
+  agents to confirm whether they are running an installed `plate` or the
+  checkout before reporting behaviour.
+- `docs/troubleshooting.md`: a symptom-keyed troubleshooting guide organised
+  around the errors the CLI emits (MQTT/FTPS connectivity, cert-pin mismatches,
+  OrcaSlicer and profile paths, camera/Docker, SSRF-blocked downloads, config
+  location). Linked from the README, the user guide, and the bug-report
+  template; ships in the sdist.
+- `tests/test_docs_links.py`: checks that cross-doc GitHub links resolve to real
+  repo paths and that every in-page anchor in the troubleshooting guide matches a
+  real heading.
+- README: a "How it compares" section describing adjacent cloud-free Bambu
+  projects and when to use them instead.
+- Printer error codes are now shown in hex as well as decimal (e.g. `Print failed
+  with error code 83935248 (hex 0x0500C010)`), and JSON error envelopes gain an
+  additive `printer_error_code_hex` field. Bambu documents these codes in hex.
+- `tests/test_docs_consistency.py`: guards against version drift in README /
+  `docs/api.md` / the bug-report template, and against `docs/manual.md` sections
+  missing from its Contents list.
+- `docs/manual.md` documents that Printables downloads are performed on the
+  user's behalf, subject to Printables' terms and each model's own licence.
+
+### Changed
+- The in-development version now carries a `.devN` suffix (`0.3.0.dev0`).
+  Previously `main` advertised the same version as the last PyPI release while
+  behaving differently — `plate --version` could not distinguish a source build
+  from the release, which made bug reports ambiguous. Drop the suffix when
+  tagging.
+- `docs/manual.md`: the OrcaSlicer section now covers per-OS installation, the
+  exact auto-detected default paths, and how to override them.
+- platecli now identifies itself honestly when talking to Printables:
+  requests carry `User-Agent: platecli/<version> (+https://github.com/DLANSAMA/platecli)`
+  instead of impersonating Chrome, and the forged `Origin`/`Referer` headers
+  are gone. Verified against the live Printables GraphQL API. Generic
+  user-supplied download URLs keep a browser-compatible User-Agent for CDN
+  compatibility, now with the honest `platecli/<version>` token appended.
+- Outbound HTTP is throttled to at most one request per second per host, and
+  `429`/`503` responses are retried while respecting `Retry-After` (clamped to
+  30 s, at most two retries).
+- README: stated the OrcaSlicer and Python 3.9+ prerequisites, qualified non-P1
+  printer support as best-effort, clarified that `plate snapshot` grabs P1/A1
+  cameras directly while X1-series needs the Docker streamer, and added
+  unattended-printing safety and maintenance-expectations sections.
+- Removed hardcoded version strings from README, `docs/api.md`, and the bug-report
+  template; the version is single-sourced from `pyproject.toml`.
+- Refreshed measured test/coverage numbers in `docs/quality-roadmap.md` and
+  `docs/test-backlog.md` (758 tests, 84.03%).
+- `AGENTS.md` retitled to the public product name.
+- `docs/api.md`: the `pause.json` / `resume.json` bullets now describe both the
+  success and the `confirmation_required` shape, matching the dual schemas.
+- CI runs weekly on a schedule (external drift: firmware, Printables API,
+  OrcaSlicer, new CVEs) and pins ruff/mypy/bandit/pip-audit versions.
+- Dependabot config added for GitHub Actions and uv dependencies.
+- New `docs/releasing.md` with the yank/rollback runbook.
+
+### Changed (BREAKING)
+- `pause` and `resume` now require `--confirm`, matching `stop`/`print`/`gcode`/`delete`.
+  Pausing mid-print parks a hot nozzle over the part; resuming an abandoned print
+  restarts motion unsupervised. Without `--confirm` they now refuse with exit code `5`
+  and (with `--json`) `"status": "confirmation_required"`. Scripts calling
+  `plate pause` / `plate resume` must add `--confirm`.
+- `print` without `--confirm` now exits `5` (`EXIT_COMMAND_ERROR`) instead of `0`.
+  The JSON payload is unchanged (`"status": "confirmation_required"`). This aligns it
+  with `stop`/`gcode`/`delete` and with the exit-code table in `docs/api.md`; a script
+  that treated exit `0` as "print started" was already silently wrong.
+
+### Fixed
+- **`scripts/bambu.py` — the documented "run from a source checkout without
+  installing" entrypoint — never worked.** Python puts the script's own
+  directory (`scripts/`) on `sys.path`, not the repo root, so it died with
+  `ModuleNotFoundError: No module named 'bambu_cli'` on any clean checkout. It
+  now prepends the repo root before importing. This mattered more than it looks:
+  with the shim broken, the only way to run the CLI was an installed `plate`,
+  which is usually an older release than the checkout — so contributors and
+  agents were reporting the *released* behaviour as if it were the source's.
+- **`scripts/clean_artifacts.py` no longer deletes the developer's `.venv`.**
+  The script appended `.venv` to its removal list whenever `GITHUB_ACTIONS` was
+  unset — i.e. on every local run — and removed it with `ignore_errors=True`. On
+  Windows any locked file (a running `python.exe`) left a *partially* deleted
+  virtualenv, after which every `uv run` failed with `No pyvenv.cfg file` and
+  exit code 106, recoverable only by hand. Nothing warned about this, and the
+  name "clean_artifacts" does not suggest destroying the dev environment. The
+  deletion also served no purpose: `tests/release_readiness_smoke.py` skips any
+  path containing `.venv`, so a present `.venv` was never flagged. `.venv` is now
+  opt-in via `--venv` / `--all`, confirmed interactively (`-y` to skip), and the
+  walk no longer descends into `.git`, `.venv`, `venv`, `.claude` or
+  `node_modules`. Removal is verified instead of best-effort: a surviving path is
+  reported with recovery steps and a non-zero exit. Adds `--dry-run`, a docstring,
+  `argparse --help`, and `tests/test_clean_artifacts.py`. CI's invocation is
+  unchanged and never passed `--venv`.
+- **`job` now validates print options on every path, not just under `--confirm`.**
+  `_run_job` only checked `--use-ams` / `--ams-mapping` when `--confirm` was passed
+  and `--upload-only` was not, so the two paths agents use as a pre-check both
+  reported success for flags that would later be rejected. `job --dry-run
+  --ams-mapping 0,1` returned `dry_run_local_skipped` even though the same flags
+  fail without `--dry-run` — inconsistent with `--infill`, which was already caught
+  in dry-run. Worse, `job --upload-only --ams-mapping 0,1` uploaded and then handed
+  back a `next_command` of `["print", …, "--ams-mapping", "0,1"]` that `print`
+  rejects with exit `5`, so an agent following the documented handoff hit a
+  guaranteed failure *after* a real upload. Validation now runs for any supplied
+  print option and still passes when none are given.
+- **`status` no longer returns a partial MQTT delta as if it were a full state
+  snapshot.** The printer publishes incremental updates on the `report` topic and
+  only sends the complete state in reply to `pushall`, but `status` returned
+  whichever message arrived first. Mid-print that was intermittently a single
+  `nozzle_temper` reading, so `plate --json status` emitted a `printer` object
+  with no `gcode_state`, `mc_percent`, or `layer_num` — a random `KeyError` for
+  any agent reading those fields. Report messages are now merged into one
+  accumulated state and `status` keeps waiting (re-issuing `pushall` on each
+  retry, honouring the existing timeout and retry count) until the state is
+  complete. If only deltas ever arrive, it now fails with a clear error naming
+  the missing keys and exit code `6` instead of emitting an incomplete object.
+  `doctor` and `print --dry-run`, which use the reply only as a liveness probe,
+  still accept the first message.
+- `status --monitor` merges deltas the same way, so a temperature-only update no
+  longer streams as `"gcode_state": "UNKNOWN"` at 0% mid-print.
+- **`setup` and `--migrate-access-code` no longer fail outright when the config
+  directory is behind filesystem virtualization.** Windows MSIX/AppContainer
+  redirection of `%APPDATA%` makes `os.replace` report `ERROR_NOT_SAME_DEVICE`
+  (`WinError 17`) even for two paths in the *same* directory, so every config and
+  access-code write raised `Could not migrate access code: [WinError 17] ...` and
+  platecli could not persist configuration at all on such a machine. The atomic
+  temp-file-plus-rename path is unchanged and still the default; only on that
+  specific errno (`EXDEV` / winerror 17) does it fall back to writing the file in
+  place, and it warns that the write is not crash-safe. The fallback writes the
+  target directly rather than copying the temp file over it, so a secret is never
+  left in a second location. Reproduced and verified against a real redirected
+  `%APPDATA%\bambu`.
+- **OrcaSlicer auto-detection now finds stock Windows installs.** The current
+  Windows installer lays the binary down as `orca-slicer.exe`, but detection only
+  probed `OrcaSlicer.exe`, so `plate setup` missed every default install and users
+  had to set `orca_slicer` by hand. Both names are now probed under each install
+  root (hyphenated first), plus a `PATH` lookup for custom locations. Found on a
+  real Windows box with OrcaSlicer 2.4.2.
+- When no OrcaSlicer exists anywhere on the machine, the `orca-slicer` /
+  `profiles-dir` errors from `preflight`, `config validate`, and `slice` now append
+  a ready-to-run install command for the host platform (`winget` / `brew` /
+  `flatpak`) instead of only suggesting a config edit — there is nothing to point
+  the config at until something is installed. When an install *is* present, the
+  existing "detected at <path>" hint still wins.
+- `tests/live_printer_smoke.py` now decodes CLI subprocess output as UTF-8 instead of
+  the platform default. On a Windows box with a cp1252 codepage the harness's reader
+  threads died with `UnicodeDecodeError` on the CLI's emoji output, losing stdout/stderr
+  while the run still reported success. `tests/agent_cli_smoke.py` already carried this
+  fix; the live harness had been missed.
+- README, AGENTS.md, SECURITY.md, docs/api.md, and docs/manual.md no longer claim a
+  universal `--confirm` gate that `pause`/`resume` did not implement; `--confirm` is
+  now documented as a deliberate-action gate rather than an authorization boundary.
+- `docs/schemas/pause.json` / `resume.json` accept the `confirmation_required` status.
+- Filenames containing square brackets (e.g. `part[v2].stl`, common in Printables model titles) no longer crash the CLI or get silently truncated in log output — the Rich log handler no longer parses log text as markup.
+- `--json` now always writes a parseable error envelope to stdout: the JSON envelope is emitted before the human-readable log line, so a logging failure can no longer leave stdout empty.
+- The envelope-before-log ordering now covers the whole package (`commands/`,
+  `camera`, `download/`, `slicer/`, `setup_cmd/`), not just `cli.py`. Every log call on
+  an error path now also routes through the shared `logging_utils.safe_log_error`, which
+  degrades to a bare stderr write instead of propagating, so a handler that raises while
+  rendering a user-controlled string can no longer swallow the `--json` envelope or
+  replace the real `BambuError` with a logging traceback.
+- `plate doctor`'s reported `camera_snapshot_note` (and `camera.py`'s snapshot
+  docstring) claimed P1P/P1S snapshots go through the BambuP1Streamer container.
+  It is the reverse: P1/A1-class cameras are captured directly with no Docker,
+  and the container is the X1-series fallback.
+- `plate preflight` now says what to do when `orca_slicer` / `profiles_dir` are not
+  configured, instead of printing `OrcaSlicer not found at` with a blank path.
+- Contributor setup uses `uv sync --extra test`, so the documented test command works
+  on a fresh clone (`uv sync` alone does not install pytest).
+- Config and access-code writes are now atomic (temp file + fsync + rename), so a crash
+  mid-write can no longer truncate `config.json` or the access-code file. `config.json`
+  also keeps a `.bak` of the previous file; access-code files are written atomically at
+  0600 but without a backup copy (fewer copies of credentials on disk). Mode stays `0600`.
+- `plate slice` now says what to do when `orca_slicer` is not configured at all, instead
+  of reporting `OrcaSlicer not found at ` with a blank path.
+- Device, file, print and G-code commands now emit the `--json` error envelope before
+  writing the human-readable log line, so a failure in the logging layer can no longer
+  leave stdout without a parseable envelope.
+- **A UTF-8 byte-order mark on `config.json` silently voided the entire config.**
+  PowerShell's `Set-Content -Encoding utf8`, Notepad's "Save as UTF-8", and most
+  Windows editors write a BOM. `json.load` then failed, the error was swallowed,
+  and every command reported `Printer IP is not configured. Please run 'plate
+  setup' first.` — advice that would have overwritten the config the user was
+  trying to repair. Config files are now read as `utf-8-sig`, which handles BOM
+  and BOM-less files identically.
+- A config that exists but fails to parse is no longer reported as *missing*.
+  `load_config` always logs the real parse error, and `plate preflight` now says
+  the file exists and could not be read rather than `Config not found at …`.
+- `plate preflight`'s Docker check is now model-aware: it no longer warns P1/A1
+  owners that "camera snapshots will be unavailable" when those printers capture
+  directly over the printer's TLS camera port. (The matching `doctor` note was
+  already corrected; this was the remaining copy of the same wrong claim.)
+- `bambu_cli/bambu.py` looks up `reconfigure` via `getattr`, fixing the sole
+  outstanding mypy error (`TextIO` has no `reconfigure`) without changing the
+  Windows encoding hardening it performs.
+- `.gitignore` now ignores `result.json`, the slice-result file OrcaSlicer's CLI
+  drops in the working directory on every `plate slice`. The existing entry read
+  `results.json` (plural) and never matched, so the file showed up as untracked
+  in every working tree after a slice.
+
+### Security
+- **`plate doctor` no longer prints your printer's LAN IP or full certificate
+  SHA-256 in its human output by default.** doctor output is routinely pasted
+  into issue reports and recorded into the README/PyPI demo GIFs. The IP now
+  reads `<redacted>`, and once a cert is pinned doctor prints a hex-free match
+  confirmation instead of the fingerprint. Pass `-v`/`--verbose` to see both.
+  The full fingerprint is still printed when the cert is *not* yet pinned (you
+  need it to pin), the MQTT-failure message still echoes the configured IP so a
+  typo is diagnosable, and the `--json` contract is unchanged.
+- The bug-report template now asks for `plate doctor --json` (which redacts the
+  LAN IP) instead of the human output.
+- All GitHub Actions are pinned to immutable commit SHAs, enforced by
+  `tests/ci_workflow_smoke.py`. Notably `pypa/gh-action-pypi-publish` was
+  tracking the mutable `release/v1` branch in the job that holds the PyPI
+  trusted-publishing `id-token: write` permission.
+- Releases now re-run the full CI matrix on the tagged commit before publishing,
+  so a tag on a red commit can no longer reach PyPI.
+- The four demo GIFs committed to the repo (`docs/demo-{dark,light}.gif`,
+  `docs/doctor-{dark,light}.gif`) previously had the maintainer's printer LAN IP
+  and full certificate SHA-256 **rasterised into the frames**, where no
+  code-level redaction could reach them. The GIFs are embedded in README.md and
+  docs/manual.md and served from raw.githubusercontent.com, making them the most
+  widely-read copy of that data. Re-recorded against the doctor redaction from
+  #61; the fingerprint now shows a hex-free match confirmation and the IP reads
+  `<redacted>`.
+
+## [0.2.2] - 2026-07-24
+
+### Added
+- `plate snapshot --unique`: timestamped output filenames
+  (`printer_snapshot_20260724T195820Z.jpg`) so repeated captures never
+  overwrite each other; composes with `--output` by inserting the stamp
+  before the extension. Designed for AI-agent workflows.
+- Snapshot `--json` output now includes `captured_at` (ISO-8601 UTC) and
+  `sha256` (digest of the JPEG bytes) so callers can verify a capture is
+  genuinely new. Additive fields; `docs/schemas/snapshot.json` updated.
+- Camera-snapshot guidance for agents in `AGENTS.md` and a Camera snapshots
+  section in `docs/manual.md`.
+
+## [0.2.1] - 2026-07-24
+
+### Fixed
+- **Slice temperature safety now covers process-section overrides**: nozzle/bed
+  temperature keys passed via `--set` or the `"process"` object of
+  `--settings-json` are validated against the same printer-safety bounds as
+  filament overrides (previously only the filament path was checked).
+- `plate doctor` honors `--network-timeout` for its MQTT, FTPS, and version
+  probes (previously hardcoded 5 s with silent retries taking ~23 s), and no
+  longer retries the MQTT check.
+- Deprecation and setup messages referenced the pre-rename `bambu setup`
+  command; they now correctly say `plate setup`.
+- The headless (non-TTY) setup error now includes a copy-pasteable
+  non-interactive example command.
+
+### Added
+- `docs/schemas/status.json`: published JSON Schema for static
+  `plate status --json` output (the monitor stream already had
+  `status_event.json`), validated in contract tests.
+- Community files: code of conduct, issue forms (bug report asks for
+  `plate doctor` output), and a pull-request template.
+
+### Changed
+- README restructured into a concise landing page; full technical
+  documentation moved to `docs/manual.md` (ships in the sdist, linked from
+  the PyPI sidebar via the new `Documentation` project URL).
+- Quickstart examples use a real Printables model (3DBenchy) instead of a
+  placeholder URL.
+- Repo hygiene: `pytest.ini` and `.coveragerc` folded into `pyproject.toml` (`[tool.pytest.ini_options]`, `[tool.coverage.*]`); removed the `.jules/` bot-notes directory. Test/coverage behavior unchanged; sdists still carry the full test config via `pyproject.toml`.
+
+## [0.2.0] - 2026-07-23
+
+### Changed
+- **Project renamed to `platecli`; the installed command is now `plate`** (was
+  `bambu-cli`, published as `bambu-local-cli`). The rename removes the vendor
+  name from the project branding; platecli remains an unofficial tool for
+  Bambu Lab printers and is not affiliated with Bambu Lab. The internal Python
+  package (`bambu_cli`), environment variables (`BAMBU_*`), and config path
+  (`~/.config/bambu/`) are unchanged, so existing configs keep working —
+  only the command you type changes. The old PyPI release `bambu-local-cli`
+  0.1.0 is yanked to point users at `platecli`.
+
+## [0.1.0] - 2026-07-18
+
+Initial development version: LAN-mode printer control (MQTT/FTPS), one-shot `job`/`send` orchestration, OrcaSlicer integration, guided setup with mDNS discovery, camera snapshots, SSRF-safe downloads with Printables support, simulation mode, and agent-facing `--json` output.
+
+### Documentation
+- Full doc truth pass (2026-07-17 audit): aligned **AGENTS.md**, **CONTRIBUTING.md**,
+  **SECURITY.md**, **README.md**, **docs/api.md**, **docs/quality-roadmap.md**,
+  **docs/test-backlog.md**, live/mutation docs. Corrected stale claims (mypy residual
+  blocklist, ≥98% coverage). Expanded config reference in README; security known
+  limitations (camera Docker bind, pin soft-fallback, HTTP downloads, pause/resume).
+  Architecture grade noted as **A−** until domain→`cli` helper extraction lands.
+  Honest metrics: ~618 non-live tests, ~82% coverage, CI floor 81, target 92 for A+/1.0.
+- **Packaging doc policy:** PyPI sdist ships only user/agent docs (`README`, `AGENTS`,
+  `SECURITY`, `CHANGELOG`, `docs/api.md`, `docs/schemas/*`). Contributor planning docs
+  (`CONTRIBUTING`, quality-roadmap, test-backlog, mutation/live-smoke) stay GitHub-only;
+  enforced by `MANIFEST.in` + package_contents forbidden list. Wheel remains runtime-only.
+
+### Security
+- **Camera streamer now binds loopback by default.** `camera_port` defaults to `127.0.0.1:1985:1984` (was `1985:1984`), so the BambuP1Streamer container no longer publishes the unauthenticated printer camera feed on all host interfaces (`0.0.0.0`). Set `camera_port` to `0.0.0.0:1985:1984` to deliberately restore LAN access. An **already-running** container keeps its old binding until recreated — run `docker rm -f bambu_camera` (the CLI now warns when it detects a running container still bound to a non-loopback interface). `camera_port` is also validated, and the localhost stream URL is now derived correctly from host-qualified specs (`[HOST:]HOSTPORT:CONTAINERPORT`).
+
+### Fixed
+- **Camera pin fail-open:** a `cert_fingerprint` **mismatch** during the direct P1/A1 camera grab now hard-aborts `snapshot` (exit 2) instead of being swallowed by the broad fallback handler and silently retried through the Docker streamer, which connects without honoring the pin. A *missing* pin still legitimately falls through to the streamer (X1 path).
+- **Camera socket fd leak:** the direct-grab TLS socket is now closed via the wrapped `SSLSocket` (which owns the fd after `wrap_socket` detaches it) rather than the detached raw socket, so a successful snapshot no longer leaks an fd / emits a `ResourceWarning`.
+- Global `--json` placed *before* the subcommand is now honored by `status`, `light`, `pause`, and `resume` (they re-declared `--json` with an implicit `False` default that clobbered the global flag, so `bambu-cli --json status` silently emitted nothing).
+- `--allow-private-ips` now actually enables private/LAN downloads for that invocation (it was parsed but never applied to runtime settings).
+- `load_access_code` and domain handlers raise structured `BambuError` / `abort` instead of calling `sys.exit` (process exit is CLI entry only).
+
+### Changed
+- Release workflow now creates the GitHub Release only after the PyPI publish succeeds (build → publish → release job chain).
+- Sdists include `pytest.ini` so the shipped test suite runs with the project's marker/coverage config.
+- Simulation mode (`--sim status`) reports representative bed/nozzle targets, fan speed, and WiFi signal instead of `?` placeholders.
+- The missing-config error now names the exact command: ``Please run `bambu-cli setup` first.``
+- `VERSION` is resolved from package metadata / `pyproject.toml` only (no duplicate string in `constants.py`).
+- MQTT status-monitor teardown no longer uses a bare `except:`.
+- Removed `@mockable` / test-awareness indirection from production code.
+- CI coverage floor enforced at **81%** (`--cov-fail-under=81`, raised from 79; multi-OS minimum — Linux ~82.3% / Windows ~81.9% branch total); single pytest path; blocking purity greps for `sys.exit` / `@mockable` / Mock branches. (Further ratchet toward the 92% A+ target remains Phase C work in `docs/quality-roadmap.md`.)
+- Package renamed to `bambu-local-cli` for PyPI publication (the `bambu-cli` name on PyPI belongs to an unrelated project). The installed command remains `bambu-cli`.
+- Wheels no longer bundle non-runtime files (`README.md`, `AGENTS.md` inside the package).
+
+### Added
+- **Full-surface slicer overrides.** `slice` (and `job`/`send`) accept `--set KEY=VALUE` / `--set-filament KEY=VALUE` (repeatable) and `--settings-json '{"process":{…},"filament":{…}}'` to override **any** of the 176 OrcaSlicer process/filament settings, not just the ~17 with dedicated flags. Unknown keys warn (with a "did you mean" suggestion) but still pass through. Temperature overrides are re-validated against the printer-safety bounds so `--set-filament nozzle_temperature=999` is refused.
+- **Slicer setting discovery:** `slice --list-settings [--json]` dumps every settable process/filament key with an example value — the agent-facing way to learn the override vocabulary.
+- Named slicer convenience flags for the common tuning knobs (sugar over the generic override machinery), each verified against a real slice: `--layer-height`, `--first-layer-height`, `--brim`, `--speed`, `--seam-position {nearest,aligned,back,random}`, `--ironing {none,top,topmost,solid}`, `--support-threshold`, `--fan-speed`, and `--flow-ratio`.
+- JSON schemas under `docs/schemas/` and contract tests in `tests/contracts/`.
+- `docs/quality-roadmap.md` scoreboard and residual coverage policy.
+- TLS pin suite (`tests/test_tls_pinning.py`); SSRF / redirect residual coverage; security + contract pytest markers.
+- PyPI trusted publishing on tagged releases (`v*`).
+- `CHANGELOG.md`, `CONTRIBUTING.md`, `SECURITY.md`, Changelog project URL.
+
+### Tests
+- CLI e2e coverage for `--allow-private-ips` wiring into `RuntimeContext` / netsafety.
+- MQTT + FTPS certificate fingerprint pin match/mismatch (and deferred-handshake) suite.
+- Expanded unit coverage for netsafety handlers, setup helpers, slicer pure paths, wizard guided flows.
+
+### Removed
+- `requirements.txt`, which duplicated the `dependencies` already declared in `pyproject.toml`. Install with `uv pip install -e .` (or `pip install platecli`) instead.
+
+[Unreleased]: https://github.com/DLANSAMA/platecli/compare/v0.5.0...HEAD
+[0.5.0]: https://github.com/DLANSAMA/platecli/compare/v0.4.0...v0.5.0
+[0.4.0]: https://github.com/DLANSAMA/platecli/compare/v0.3.0...v0.4.0
+[0.3.0]: https://github.com/DLANSAMA/platecli/compare/v0.2.2...v0.3.0
+[0.2.2]: https://github.com/DLANSAMA/platecli/compare/v0.2.1...v0.2.2
+[0.2.1]: https://github.com/DLANSAMA/platecli/compare/v0.2.0...v0.2.1
+[0.2.0]: https://github.com/DLANSAMA/platecli/compare/v0.1.0...v0.2.0
+[0.1.0]: https://github.com/DLANSAMA/platecli/releases/tag/v0.1.0