Repository navigation
docs: clarify inert context windows and remove stale ToolConfig entry - #198
Conversation
Dz-Task-Id: tsk_30efb5d234a615250fe87790926f7de9
|
State that context_windows is inert, and drop the stale ToolConfig docstring bullet
Work landed on #198 — watching it toward merge. 7m · 24 tool calls · 4 messages Sessions: coding run 🤖 developerz.ai, an automated agent, not a human — hv-dedicated-1 (developerz-ai[bot]) · model |
✅ Reviewed — nothing blocking0 actionable comment(s) · grounded on your code ⏱ 49s wall clock · 🤖 developerz.ai — automated review, running on your model and your box. |
There was a problem hiding this comment.
Review summary — 5 file(s), 0 finding(s).
Clarifies that context-window settings are inert metadata and removes the stale ToolConfig inventory entry without changing runtime behavior. No blocking defects found in the supplied diff; review completed .
- test files were not shown to the reviewer — this diff touched no path recognised as a test, so no assertion was available to read the change against
🤖 developerz.ai review — automated, running on your model and your box. What is this? · Reviewed by custom/gpt-6.1-sol on box hv-dedicated-1 — 129.5k in / 363 out over 3 calls, on your flat-rate plan: no per-token charge (~$0.26 at OpenRouter list rates).
…rop-the--30efb5d2
There was a problem hiding this comment.
Review summary — 5 file(s), 0 finding(s).
Completed review documentation now identifies context-window settings as inert metadata and removes the stale ToolConfig entry. Runtime values, logic, and test assertions are unchanged; no blocking defects found in the supplied diff.
- test files were not shown to the reviewer — this diff touched no path recognised as a test, so no assertion was available to read the change against
🤖 developerz.ai review — automated, running on your model and your box. What is this? · Reviewed by custom/gpt-6.1-sol on box hv-dedicated-ci-0 — 129.5k in / 295 out over 3 calls, on your flat-rate plan: no per-token charge (~$0.26 at OpenRouter list rates).
## What
The repo-plan endpoint docs named two conflicting hard-coded lists
(Read, Glob, Grep vs Read, Glob, Grep, Bash). The real list comes from
config.tools.planning via get_tools_for_phase('planning'). All five
sites
now say so instead of hard-coding a list.
## Acceptance
- api/models_repo.py, api/routes_repo.py, mcp/server.py,
mcp/tool_handlers_repo.py and docs/mcp-tools.md describe the planning
tools as configured via config.tools.planning
- No hard-coded '(Read, Glob, Grep)' or '(Read, Glob, Grep, Bash)' list
remains at those five sites
- bin/check is green
## Changes
- `src/claude_task_master/api/models_repo.py` (edit) — PlanRepoRequest
docstring (lines 122-124): replace hard-coded '(Read, Glob, Grep)' with
wording that the read-only planning tools come from
config.tools.planning (via get_tools_for_phase('planning')). Docstring
only; keep lines <=100 cols.
- `src/claude_task_master/api/routes_repo.py` (edit) — post_plan_repo
docstring (line 275): replace '(Read, Glob, Grep)' with 'the planning
tools configured in config.tools.planning'. Also optionally align the
summary description string at line 267 ('Uses read-only tools') to
mention the configured planning tools. Text only.
- `src/claude_task_master/mcp/server.py` (edit) — plan_repo tool
docstring (lines 263-264): replace '(Read, Glob, Grep)' with the
planning tools configured in config.tools.planning
(get_tools_for_phase('planning')). Docstring is the MCP tool
description; keep it plain text.
- `src/claude_task_master/mcp/tool_handlers_repo.py` (edit) — plan_repo
handler docstring (lines 313-314): replace '(Read, Glob, Grep, Bash)'
with the planning tools configured in config.tools.planning
(get_tools_for_phase('planning')). Use the same phrasing as
mcp/server.py.
- `docs/mcp-tools.md` (edit) — Line 654: replace 'read-only tools (Read,
Glob, Grep, Bash)' with 'the planning tools configured in
`config.tools.planning` (default: Read, Glob, Grep, WebFetch,
WebSearch)'. Default per core/config.py:168-171. Must match phrasing
used in the MCP docstrings.
- `docs/repo-setup.md` (edit) — [docs] Lines 265 and 271-276 describe
the same repo-plan endpoint with a hard-coded 'read-only tools (Read,
Glob, Grep, Bash)' list and a 'Read-Only Tools Used' list that includes
Bash. docs/mcp-tools.md:654 now says the tools come from
config.tools.planning (default Read, Glob, Grep, WebFetch, WebSearch),
so a reader sees two conflicting lists. Reword line 265 to say it uses
the planning tools configured in `config.tools.planning`. Retitle the
list at 271-276 as the default planning tools: Read, Glob, Grep,
WebFetch, WebSearch. Drop the Bash bullet and add WebFetch and WebSearch
bullets.
## Verification
`./bin/check` passed (exit 0) in 7m 18s.
What the editors ran to see the change working, and what they saw:
- `python verify_docs_links.py` (docs): All 299 links verified
successfully!
- `uv run ruff check on both files, then uv run pytest tests/mcp -q
--no-cov` (mcp): Ruff: All checks passed. Pytest: 1 failed, 247 passed.
The failed test changed between runs (first
test_setup_repo_nodejs_project, then
test_setup_repo_with_multiple_project_types), both in
test_tools_repo_setup.py. The trace ends in selectors.py:398: Failed,
which looks like a timeout in the setup-repo tests. I did not diagnose
it. A rerun of test_setup_repo_nodejs_project on its own printed no
Failed, Timeout or Error lines.
- `uv run ruff check on both files` (api): All checks passed!
- `uv run pytest tests/api -q -p no:cacheprovider --no-cov (the scoped
command)` (api): Not green as run. The 2s per-test timeout fired on cold
imports: first run stopped after 1 failure, the second showed '207
passed, 40 errors'.
- `uv run pytest tests/api -q -p no:cacheprovider --no-cov --timeout=60`
(api): 247 passed in 52.65s. The edits are docstring and description
text only.
Part 2/7 of goal `rtk_3f7461e0031b4617af6a313e24e563cc`
- #196
- #199
- #197
- developerz-ai/claude-task-master (no PR yet)
- #198
- developerz-ai/claude-task-master (no PR yet)
<!-- dz-task:tsk_1a500a911a427cb0273167dfc07eb704 -->
---
🤖 Opened by **developerz.ai** for task
`tsk_1a500a911a427cb0273167dfc07eb704` on
developerz-ai/claude-task-master.
Raised automatically by the `raw_task_triage` lane.
<!-- dz-stack:start -->
> 🤖 developerz.ai, an automated agent, not a human — this PR is one
slice of a stacked plan.
> Dispatched automatically by the `raw_task_triage` lane
> Model `claude-sonnet-5-5` on `router.ai.developerz.ai` via `native`,
funded by `account_key`
> Ran on box `hv-dedicated-ci-0`
> Cost unknown: no list price on record for `claude-sonnet-5-5` on
`router.ai.developerz.ai`
> Run timeline
[`run_56326b2d8ea34a858a823fd4f3f8acef`](https://app.developerz.ai/fleet/runs/run_56326b2d8ea34a858a823fd4f3f8acef)
> Part 2/7 of plan
[`rtk_3f7461e0031b4617af6a313e24e563cc`](https://app.developerz.ai/fleet/plans/rtk_3f7461e0031b4617af6a313e24e563cc)
<!-- dz-stack:end -->
Co-authored-by: developerz-ai[bot] <286432749+developerz-ai[bot]@users.noreply.github.com>
## Acceptance - With all loader-mapped overrides unset, claudetm config show --env lists every CLAUDETM_MODEL_* entry from ENV_VAR_MAPPINGS, including FABLE and SONNET_1M with their correct config paths. - The CLI regression derives expected model variables from loader mappings and fails if either missing table row is removed; existing active-override and secret-masking tests pass. - ./bin/check exits successfully after the change. ## Changes - `tests/cli/test_config_commands.py` (edit) — Strengthen test_config_show_env_no_overrides: clear every ENV_VAR_MAPPINGS key, invoke config show --env with a wide test console, and assert the rendered model variable names and config paths match the loader mappings filtered by CLAUDETM_MODEL_. Replace the stale manual clearing list; the test must fail before the table fix. - `src/claude_task_master/cli_commands/config.py` (edit) — Add available-variable table rows for CLAUDETM_MODEL_FABLE → models.fable and CLAUDETM_MODEL_SONNET_1M → models.sonnet_1m, with model-name descriptions. Preserve the current no-overrides/active-overrides branches and secret masking; no loader or API changes. ## Verification `./bin/check` passed (exit 0) in 9m 09s. What the editors ran to see the change working, and what they saw: - `uv run pytest tests/cli/test_config_commands.py -q` (config-env): Before table fix: 5 failed, 27 passed. Strengthened regression reported missing CLAUDETM_MODEL_FABLE → models.fable and CLAUDETM_MODEL_SONNET_1M → models.sonnet_1m. Four unrelated TestConfigPath failures showed Rich wrapping .claude-task-master across lines. - `uv run pytest tests/cli/test_config_commands.py -q` (config-env): After table fix: 4 failed, 28 passed. Model-mapping regression, active-override tests, and secret-masking tests passed. Remaining failures: test_config_path_shows_path, test_config_path_output_is_pipeable, test_config_path_check_file_exists, test_config_path_check_file_not_exists; output contained '.claude-task\n-master/config.json'. Exit 1; full gate not run. ## Notes Premises of the request the team disproved: - (from a dropped noWork claim) None; the request is valid. Part 7/7 of goal `rtk_3f7461e0031b4617af6a313e24e563cc` - #196 - developerz-ai/claude-task-master (no PR yet) - #199 - #197 - #201 - #198 <!-- dz-task:tsk_0b39fea6c7f8a90c5e86acaaadcb2efa --> --- 🤖 Opened by **developerz.ai** for task `tsk_0b39fea6c7f8a90c5e86acaaadcb2efa` on developerz-ai/claude-task-master. Raised automatically by the `raw_task_triage` lane. <!-- dz-stack:start --> > 🤖 developerz.ai, an automated agent, not a human — this PR is one slice of a stacked plan. > Dispatched automatically by the `raw_task_triage` lane > Model `gpt-6-astra` on `router.ai.developerz.ai` via `native`, funded by `account_key` > Ran on box `hv-dedicated-dev-marshall` > Cost unknown: no list price on record for `gpt-6-astra` on `router.ai.developerz.ai` > Run timeline [`run_a22fd4369ef447b195caee2b90774cf7`](https://app.developerz.ai/fleet/runs/run_a22fd4369ef447b195caee2b90774cf7) > Part 7/7 of plan [`rtk_3f7461e0031b4617af6a313e24e563cc`](https://app.developerz.ai/fleet/plans/rtk_3f7461e0031b4617af6a313e24e563cc) <!-- dz-stack:end --> Co-authored-by: developerz-ai[bot] <286432749+developerz-ai[bot]@users.noreply.github.com>
Acceptance
Changes
README.md(edit) — Correct context-window guidance at lines 85-109, 315, 345-360 and 377: distinguish working model overrides from currently inert context_windows/CLAUDETM_CONTEXT_* metadata. Replace the auto-compaction note and remove advice to tune these values for account tiers or gateways; it cannot affect runtime compaction or grant context. Preserve config keys/examples and unrelated model/CLI documentation.src/claude_task_master/core/config.py(edit) — Rewrite ContextWindowsConfig documentation at lines 119-138 to state that values are retained/loadable but currently inert: no runtime compaction threshold or larger SDK window results. Align its field descriptions and the root context_windows description at line 254, removing actionable tier/gateway tuning advice. Preserve all defaults, types, keys and runtime logic.src/claude_task_master/core/agent_models.py(edit) — Remove the nonexistent ToolConfig bullet at line 7. Replace lines 107-112 with accurate inert-context metadata commentary; clarify get_context_window's docstring and the unused threshold comments at lines 214-215 so none promises runtime compaction. Preserve constants, helper implementation and exports; the remaining module inventory must name actual definitions.src/claude_task_master/core/profiles.py(edit) — Replace lines 238-244 with the actual contract: profile context values are emitted as CLAUDETM_CONTEXT_* and loaded into config, using the existing family fallback, but are currently inert for runtime compaction/window selection. Clarify related context_windows documentation as needed without changing environment emission, fallback rules or profile persistence.tests/core/test_profiles.py(edit) — Rewrite test_context_family_fallback_covers_all_tiers's docstring at lines 256-259 to describe the asserted sonnet_1m/fable environment-value inheritance, noting that context_windows remains inert for compaction. Keep the existing test inputs and assertions unchanged; the test proves propagation, not prevention of context overflow.Verification
./bin/checkpassed (exit 0) in 3m 17s.What the editors ran to see the change working, and what they saw:
python verify_docs_links.py(docs): Exit code 0; scanned 49 Markdown files and reported: All 299 links verified successfully!uv run ruff check src/claude_task_master/core/config.py src/claude_task_master/core/agent_models.py src/claude_task_master/core/profiles.py tests/core/test_profiles.py(python-docs): Exit code 0; All checks passed!Notes
Premises of the request the team disproved:
Part 6/7 of goal
rtk_3f7461e0031b4617af6a313e24e563cc🤖 Opened by developerz.ai for task
tsk_30efb5d234a615250fe87790926f7de9on developerz-ai/claude-task-master.Raised automatically by the
raw_task_triagelane.