feat(method): quality-floors — 4 zero-turn input-dialect floors (wm2 miss → mechanical checks)#144
Closed
pilotspacex-byte wants to merge 4 commits into
Closed
feat(method): quality-floors — 4 zero-turn input-dialect floors (wm2 miss → mechanical checks)#144pilotspacex-byte wants to merge 4 commits into
pilotspacex-byte wants to merge 4 commits into
Conversation
…act's own example formats (quality-floors 1/4) Closes the WV1 wm2 failure CLASS for ~zero turns (benchmark evidence: an arm's own suite stayed green on naive timestamps while the spec's examples were Z-suffixed — the aware/naive crash shipped green at 0.80): - add_engine/constants.py: _DIALECT_CLASSES closed registry (v1: one class, aware-iso-timestamp; full-shape regex — bare dates never match), named in __all__ like _GATE_MODES. - add.py: _dialect_gaps PURE helper (frozen section-3 body vs declared section-4 test files, fail-open) + a print-only tests->build crossing warning naming the class and the re-cross repair (never refuses; exit/state byte-identical) + a dialect_gap WARN in check over ACTIVE build/verify/done tasks (measure-not-block). - 6 red-first pins (test_spec_dialect_floor.py, live-CLI board harness); three trees byte-identical; ENGINE_MD5 7f96609e / PKG 710a009f; SEAMS x17 re-pin. Full tooling suite 3367 green. Tripwire lesson recorded in-task: a bare directory token in section-4 "Tests live in:" sweeps add.py itself into the tamper snapshot — the build's own target must never be declared as a test file (honest redo via TESTS re-cross, attempt 1). Task: spec-dialect-floor (oneshot lane, AI-plan-verify freeze @v1), milestone quality-floors. author: Tin Dang
…lary (quality-floors 3+4) Task dialect-check-and-data-vocab (quality-floors levers 3+4, gate PASS, frozen @v1 via ai-plan-verify). Two more zero-turn floors closing the wm2 input-dialect class from the WV1 benchmark: - TASK.md.tmpl `### Deep checks` gains the DIALECT line ("tests speak the same value formats the spec's examples use") — audit-counted by the existing `shallow` machinery, zero engine code; TASK.fast.md.tmpl §6 gains the condensed twin ("input dialect held"). - GLOSSARY.md.tmpl `## Sensitivity classes` + this project's .add/GLOSSARY.md + sensitivity.md skill guide gain the PROSE rule: datetime, money, or timezone arithmetic ⇒ declare `data` (full lane), wm2 evidence cited. No new sensitivity token minted — _project_sensitivity_domain proven () over a fresh render. - Byte fences held by compression, not rebaseline: TASK.md.tmpl 11391B (≤11400), skill pool/tree back under 51885/145974 via ~183B trimmed from sensitivity.md prose. - 5 sibling pins amended via an explicit TESTS re-cross, as the frozen least-sure flag predicted: 4 Deep-checks fixtures now fill the DIALECT placeholder (guarantee-lints, advisor-review-step, advisor-verdict-audit, verify-record-rollup); fast-lane-skips' task-local empty-git-diff guard re-pinned to its durable invariant (skip machinery stays fast-lane-only). - New red suite test_dialect_vocab_lines.py (6 pins). All twins synced incl. the 4th gitignored dogfood tree. Full suite 3374 passed. - §6 OBSERVE: lever-1 false positive found dogfooding — _dialect_gaps scans §3 raw body incl. the engine's own freeze timestamp; left as an open SPEC delta. refs: quality-floors milestone (task 2/3) · benchmark/results/2026-07-wv1-rep0.md author: Tin Dang
… input-format boundary (quality-floors 2) Task fast-lane-boundary-line (quality-floors lever 2, gate PASS, frozen @v1 via ai-plan-verify). The last of the 4 zero-turn floors closing the WV1 wm2 input-dialect class: - TASK.fast.md.tmpl §1 gains, directly after Accept:, a `Boundary:` line — >=1 format-variant per external input shape the tests must speak (e.g. aware vs naive timestamp), or an explicit "none — no external input". - cmd_freeze refuses a task whose Boundary value is still the bare template placeholder (or empty): `boundary_unfilled`, fired in the shared validate block directly after unflagged_freeze — BOTH freeze paths (human and --ai-plan-verify), validate-then-write (zero bytes land on refusal). A task with NO Boundary: line is grandfathered: legacy fast tasks and the full lane gain no new refusal. Placeholder rule = empty or bare unfenced `<...>`, backtick-exempt (mirrors _section_unfilled). - New red suite test_fast_boundary_line.py (9 pins, live-CLI board: refusal exit codes + byte-identical state on refusal + both paths + grandfathering + §3-fence text never triggers). - 2 test_fast_lane_skips fixtures amended via an explicit TESTS re-cross (they freeze fresh-rendered fast tasks, which now scaffold the placeholder line): shared _fill_boundary helper. - ENGINE_MD5 re-aimed 16cd7cca (prior 7f96609e); SEAMS scope-token-grammar pin x18 (:5337); add.py + fast template synced across all twins incl. the 4th dogfood tree. Full suite 3383 passed. refs: quality-floors milestone (task 3/3) · benchmark/results/2026-07-wv1-rep0.md author: Tin Dang
…hipped) All 4 human-picked input-dialect floors from the WV1 wm2 miss are live and gated PASS: 1. spec-dialect floor (f9d2303) — tests->build dialect_gap warning 2. fast-lane Boundary line (b2483be) — freeze refuses boundary_unfilled 3. §6 dialect check line (bb72452) — Deep-checks DIALECT bullet 4. datetime/money/tz => data (bb72452) — sensitivity vocabulary milestone-done wrote RETRO.md; archive-milestone removed it from active state (files on disk untouched). Balance principle held: every floor is a ~zero-turn mechanical check at an existing seam — the lean lane keeps its cost, the failure class is closed without re-added ceremony. One SPEC delta left open (task 2 gate record): lever 1's _dialect_gaps scans the §3 raw body incl. the engine's own freeze timestamp, a self-referential false positive found dogfooding — candidate future task. refs: quality-floors milestone author: Tin Dang
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
Comment on lines
+89
to
+90
| "Least-sure flag surfaced at freeze: [contract] narrow " | ||
| "shape — cost: one re-freeze.", |
| import unittest | ||
| from pathlib import Path | ||
|
|
||
| import add |
| Run: python3 -m unittest test_fast_boundary_line -v | ||
| """ | ||
| import io | ||
| import json |
4 tasks
Contributor
Author
|
Superseded by #145, which includes these exact quality-floors commits (f9d2303, bb72452, b2483be, ad950c2) verbatim as the base of a 3-milestone branch (quality-floors + expectations-first + plan-legibility). Reviewing/merging #145 lands quality-floors along with the two later milestones. Closing to avoid duplicate/stacked-PR conflicts. |
pilotspacex-byte
added a commit
that referenced
this pull request
Jul 13, 2026
…ty-floors (3 milestones, supersedes #144) (#145) * feat(method): spec-dialect floor — the red suite must speak the contract's own example formats (quality-floors 1/4) Closes the WV1 wm2 failure CLASS for ~zero turns (benchmark evidence: an arm's own suite stayed green on naive timestamps while the spec's examples were Z-suffixed — the aware/naive crash shipped green at 0.80): - add_engine/constants.py: _DIALECT_CLASSES closed registry (v1: one class, aware-iso-timestamp; full-shape regex — bare dates never match), named in __all__ like _GATE_MODES. - add.py: _dialect_gaps PURE helper (frozen section-3 body vs declared section-4 test files, fail-open) + a print-only tests->build crossing warning naming the class and the re-cross repair (never refuses; exit/state byte-identical) + a dialect_gap WARN in check over ACTIVE build/verify/done tasks (measure-not-block). - 6 red-first pins (test_spec_dialect_floor.py, live-CLI board harness); three trees byte-identical; ENGINE_MD5 7f96609e / PKG 710a009f; SEAMS x17 re-pin. Full tooling suite 3367 green. Tripwire lesson recorded in-task: a bare directory token in section-4 "Tests live in:" sweeps add.py itself into the tamper snapshot — the build's own target must never be declared as a test file (honest redo via TESTS re-cross, attempt 1). Task: spec-dialect-floor (oneshot lane, AI-plan-verify freeze @v1), milestone quality-floors. author: Tin Dang * feat(method): §6 dialect check line + datetime/money/tz ⇒ data vocabulary (quality-floors 3+4) Task dialect-check-and-data-vocab (quality-floors levers 3+4, gate PASS, frozen @v1 via ai-plan-verify). Two more zero-turn floors closing the wm2 input-dialect class from the WV1 benchmark: - TASK.md.tmpl `### Deep checks` gains the DIALECT line ("tests speak the same value formats the spec's examples use") — audit-counted by the existing `shallow` machinery, zero engine code; TASK.fast.md.tmpl §6 gains the condensed twin ("input dialect held"). - GLOSSARY.md.tmpl `## Sensitivity classes` + this project's .add/GLOSSARY.md + sensitivity.md skill guide gain the PROSE rule: datetime, money, or timezone arithmetic ⇒ declare `data` (full lane), wm2 evidence cited. No new sensitivity token minted — _project_sensitivity_domain proven () over a fresh render. - Byte fences held by compression, not rebaseline: TASK.md.tmpl 11391B (≤11400), skill pool/tree back under 51885/145974 via ~183B trimmed from sensitivity.md prose. - 5 sibling pins amended via an explicit TESTS re-cross, as the frozen least-sure flag predicted: 4 Deep-checks fixtures now fill the DIALECT placeholder (guarantee-lints, advisor-review-step, advisor-verdict-audit, verify-record-rollup); fast-lane-skips' task-local empty-git-diff guard re-pinned to its durable invariant (skip machinery stays fast-lane-only). - New red suite test_dialect_vocab_lines.py (6 pins). All twins synced incl. the 4th gitignored dogfood tree. Full suite 3374 passed. - §6 OBSERVE: lever-1 false positive found dogfooding — _dialect_gaps scans §3 raw body incl. the engine's own freeze timestamp; left as an open SPEC delta. refs: quality-floors milestone (task 2/3) · benchmark/results/2026-07-wv1-rep0.md author: Tin Dang * feat(method): fast-lane Boundary: line — freeze refuses an undeclared input-format boundary (quality-floors 2) Task fast-lane-boundary-line (quality-floors lever 2, gate PASS, frozen @v1 via ai-plan-verify). The last of the 4 zero-turn floors closing the WV1 wm2 input-dialect class: - TASK.fast.md.tmpl §1 gains, directly after Accept:, a `Boundary:` line — >=1 format-variant per external input shape the tests must speak (e.g. aware vs naive timestamp), or an explicit "none — no external input". - cmd_freeze refuses a task whose Boundary value is still the bare template placeholder (or empty): `boundary_unfilled`, fired in the shared validate block directly after unflagged_freeze — BOTH freeze paths (human and --ai-plan-verify), validate-then-write (zero bytes land on refusal). A task with NO Boundary: line is grandfathered: legacy fast tasks and the full lane gain no new refusal. Placeholder rule = empty or bare unfenced `<...>`, backtick-exempt (mirrors _section_unfilled). - New red suite test_fast_boundary_line.py (9 pins, live-CLI board: refusal exit codes + byte-identical state on refusal + both paths + grandfathering + §3-fence text never triggers). - 2 test_fast_lane_skips fixtures amended via an explicit TESTS re-cross (they freeze fresh-rendered fast tasks, which now scaffold the placeholder line): shared _fill_boundary helper. - ENGINE_MD5 re-aimed 16cd7cca (prior 7f96609e); SEAMS scope-token-grammar pin x18 (:5337); add.py + fast template synced across all twins incl. the 4th dogfood tree. Full suite 3383 passed. refs: quality-floors milestone (task 3/3) · benchmark/results/2026-07-wv1-rep0.md author: Tin Dang * chore(method): close + archive quality-floors milestone (3/3 floors shipped) All 4 human-picked input-dialect floors from the WV1 wm2 miss are live and gated PASS: 1. spec-dialect floor (f9d2303) — tests->build dialect_gap warning 2. fast-lane Boundary line (b2483be) — freeze refuses boundary_unfilled 3. §6 dialect check line (bb72452) — Deep-checks DIALECT bullet 4. datetime/money/tz => data (bb72452) — sensitivity vocabulary milestone-done wrote RETRO.md; archive-milestone removed it from active state (files on disk untouched). Balance principle held: every floor is a ~zero-turn mechanical check at an existing seam — the lean lane keeps its cost, the failure class is closed without re-added ceremony. One SPEC delta left open (task 2 gate record): lever 1's _dialect_gaps scans the §3 raw body incl. the engine's own freeze timestamp, a self-referential false positive found dogfooding — candidate future task. refs: quality-floors milestone author: Tin Dang * feat(method): expectations-first flow — collapse ground+contract into the plan phase (engine core) Reorder the task lifecycle to expectations-first. The 9-phase flow (ground→specify→scenarios→contract→tests→build→verify→observe→done) becomes an 8-phase flow (specify→scenarios→plan→tests→build→verify→observe→done): grounding, the frozen contract, and the build strategy unify into ONE `plan` phase carrying the single human freeze. Specify/scenarios become light projections of a milestone-level Ground + the request — no re-grounding per task. Engine core (this commit; the existing-test migration follows): - constants.py: PHASES/PHASE_GROUPS/PHASE_OWNER/PHASE_AGENT/PHASE_GUIDE re-pointed; DIRECTION = (specify,scenarios,plan,tests); plan owner = seam; fallback templates. - add.py: freeze allowed at >= plan; the freeze gate MOVES from tests->build to the plan->tests crossing (the red suite is written FROM the frozen contract), with a recorded --skip-freeze that carries through _build_entry; _grounded_state / ground anchor-SHA / scope-lock read §3 PLAN sub-blocks; _FRONT_PHASES names plan; decide-seam label PLAN; legacy load maps ground->specify, contract->plan (fail-soft _phase_index + idempotent _normalize_phase_tokens on read). The cross-component contract system is UNTOUCHED (only the phase token changed). - templates: §0 GROUND removed; §3 PLAN carries ### Grounding / ### Contract (first fence, tamper-HARD) / ### Build-strategy (SOFT); §5 BUILD is execution-only. - engine_pin: ENGINE_MD5 re-aimed to 07659b10; 3-tree byte parity held. - test_plan_phase_flow.py: new red/green suite (16 tests) pinning the whole change. author: Tin Dang <tindang.ht97@gmail.com> * test(method): migrate the full suite to the expectations-first plan phase + harden the engine Second half of plan-phase-core: migrate every existing test to the new 8-phase flow (specify→scenarios→plan→tests→build→verify→observe→done) and fix the engine defects the migration surfaced. 274 red → green with no weakened assertion; 3399 tests pass. Test migration (61 files, 8 themed batches): - phase tokens (ground→specify, contract→plan), advance counts (specify→build = 4), and the §-heading renames (§0 GROUND gone; §3 CONTRACT→PLAN with Grounding/Contract/Build-strategy sub-blocks) re-pointed across grounding, freeze-gate, fast-lane, cross-component, bundle/advance, template, ai-plan-verify, and strategy/persona suites. Fixtures re-pointed to the new locations; facts genuinely removed updated to new-truth; the cross-component contract SYSTEM stayed untouched. Engine hardening (defects the 9→8 PHASES shrink introduced, all found by the migration): - task_phases / advance --fill / status --section used the phase ordinal as the literal §-number; now map ordinal→section (n+1) since §0 is gone. Fixes the phase-detail off-by-one + IndexError. - render_task_detail's hardcoded `p["n"] == 6` verify branch → keys on phase name. - ADR facet harvest read §5; the facets moved to §3 ### Build-strategy → reads §3. - _build_entry's freeze-skip carry-through leaked into re-cross (a DRAFT §3 could re-cross); a new require_frozen flag makes re-cross re-assert the freeze — "re-cross is never a freeze bypass" holds. - skip-rationale read the dead §0; now reads the task header (where `skips:` lives) — advance + audit. - _FALLBACK_TASK_FAST regained its ## 5 · BUILD heading; the fast template regained a Skip rationale line; the template §3 regained the "never retro-red" grandfather clause. Prose ripple (kept the suite green; deep book rewrite stays T4): - ch02 flow diagram + prose (×4 twins) + diagrams/CHECKLIST name the plan phase (grounding folds into Plan, not a §0 preamble); the interactive installer showcase (cli.js / _installer.py) names Plan not Contract; .add/SEAMS.md anchor re-pinned to _declared_scope's current line. Trees: 4-way engine/template parity (tooling · _bundled · .add · add-method/.add); bundle docs synced; ENGINE_MD5→33f46b7d, ENGINE_PKG_MD5→a968f9e6 re-pinned. author: Tin Dang <tindang.ht97@gmail.com> * chore(method): record plan-phase-core verify gate PASS (T1 expectations-first) Close T1's verify phase for the 9->8 expectations-first flow collapse. Evidence (all confirmed by direct observation, not test names alone): - full suite 3397/3397 OK; guard set 51/51 OK - fresh task seeds phase specify; walk reaches plan before tests - rendered TASK.md has no §0 GROUND; §3 PLAN carries Grounding/Contract/Build-strategy - legacy ground/contract states load without crash (ground->specify migration; 0 tracebacks) - engine byte-parity x3 (add.py) / x4 (templates) == pins; cross-component contract SYSTEM untouched Refute-read EARNED: audited the whole migration diff — net +1 assertion (146/145), 0 assertRaises dropped, the 4 removed strong asserts are §0-removal index re-points, all 10 deleted test methods have 1:1 8-phase replacements. Coverage increased. Advisor 3-lens PASS: Security CLEAR (freeze gate moved plan->tests, preserved; re-cross re-asserts it), Concurrency CLEAR, Architecture PASS with one disclosed residue — deep skill/glossary/book prose still names a "Contract" phase, deferred to T4 book-align. Gate crossing tripped the post-freeze test-change tripwire (the 61-file migration is this task's TESTS-phase deliverable, fingerprinted at the earlier tests->build crossing); re-crossed tests->build (human-approved, Tin Dang) to re-snapshot the migrated tree, then recorded PASS honestly. Reviewed by Tin Dang, 2026-07-12. author: Tin Dang <tindang.ht97@gmail.com> * feat(method): milestone-level `## Ground` seed — specify projects from it (T2 expectations-first) Second task of expectations-first: a milestone gains a thin `## Ground` section — shared real-code context gathered ONCE — that each task's `specify` projects its §1 expectations from, so tasks stop re-grounding shared context. Prose/template-only; the engine digest is unchanged (a heavy milestone-ground validator stays deferred, per the milestone Out). What shipped: - MILESTONE.md.tmpl (×3 trees) carries `## Ground` right after `## Scope`, before `## Shared decisions`, with four milestone-altitude fields: Touches · Anchors · Honors (conventions) · Issues/Risks (shared). A fresh `new-milestone` scaffolds it. - phases/1-specify.md (×3) Converge step now cues: draft §1 by PROJECTING from the milestone `## Ground` + the request (replacing the stale §0-GROUND reference). - scope.md (×3) Position-the-goal step records the goal-grounding AS the milestone's `## Ground` section — it persists now instead of being discarded once the goal is written. - test_milestone_ground.py (new) pins presence/placement/thinness/specify-cue/scope-mention/ byte-parity/engine-untouched; test_scope_loop bumped 9→10 named sections (adds `ground`). Two engineering notes worth recording: - LEAN BUDGET, NOT BUMP: the added prose pushed the phases pool, the reference pool, and the whole-skill-tree byte budgets over. Absorbed by COMPRESSING existing verbose prose under the existing frozen targets (33284 / 51885 / 145974) — never by raising a target. - CHANGE REQUEST, NOT SILENT EDIT: the v1 field label `Conventions/Seams:` collided with the ubiquitous-language slang guard (bans bare 'seam' on the template surface) — missed at grounding. Corrected honestly: phase plan → rename to `Honors (conventions):` → re-freeze @ v2 (human-approved) → re-cross tests → rebuild. The frozen contract was never edited to pass a build. Verify: full suite 3408/3408 OK; refute-read EARNED; advisor 3-lens PASS (one disclosed prose residue — scope.md goal-grounding fields vs the `## Ground` fields diverge slightly → T3). Reviewed by Tin Dang, 2026-07-12. author: Tin Dang <tindang.ht97@gmail.com> * feat(method): unify the plan-phase guide + realign SKILL/book to 8 phases (guides-and-skill, T3 expectations-first) Third task of expectations-first: retire the two split phase guides — `phases/0-ground.md` (grounding) + `phases/3-contract.md` (freeze the shape) — into ONE unified `phases/3-plan.md` that teaches the plan phase end-to-end (ground the real code · freeze the contract · set the build strategy), and cascade the rename through SKILL.md, the book, and the engine's ref. What shipped: - skill (×3 trees, byte-identical): NEW `phases/3-plan.md` merges the two retired guides — Grounding (Touches·Context·Honors·Anchors·Issues/Risks·Related-intent·Ground-SHA + the gather-method hint) → Contract (freeze the external shape, a response per Reject code) → Build-strategy; keeps the 7-item freeze-review checklist and the v16 `<output_format>`/ `<prompt>`/`<exit_gate>` XML convention. `0-ground.md` + `3-contract.md` DELETED. - SKILL.md (×3): phase table drops the `ground` + `contract` rows for one `plan` row → `phases/3-plan.md`; flow prose reads specify→scenarios→plan→tests→build→verify→observe. Dangling refs healed in scope.md · run.md · components.md · fast-lane.md · 2-scenarios.md. - book (×4 trees): `git mv 05-step-3-contract.md → 05-step-3-plan.md`; the `05-step-3-contract` PATH survives NOWHERE (mkdocs nav + 03/04/06/10/appendix-f/README/GETTING-STARTED/agents cross-links all follow); the glossary "Ground" entry's guide path re-pointed (deep term NARRATIVE deferred to T4). TASK.md.tmpl §3 heading (×4 twins) points at the new chapter. - engine: `add_engine/constants.py` PHASE_GUIDE["plan"] chapter → `05-step-3-plan.md`; ENGINE_PKG_MD5 re-aimed a968f9e6 → 28212a55; add.py UNCHANGED (ENGINE_MD5 stable). - tests: NEW `test_guides_skill_realigned.py` grep-guard (guide existence/deletion · SKILL table · engine ref · zero stale chapter path · no stale phase prose · dangling-guide refs). ~21 existing suites re-pointed (pool-lists · freeze-checklist/ground-content constants · counts 32→31 · a 4th template twin) — every content FLOOR preserved in 3-plan.md, never a weakened assertion. Engineering notes worth recording: - BLAST RADIUS AS FLAGGED: the §1 ⚠ (guide-rename ripple across enumerated test lists) realized exactly — ~21 files, all mechanical re-point, same class as plan-phase-core's 78-file migration. Content floors (7-item checklist · four grounding fields · completeness rubric · context categories · gather-method hint · SHAPE/unflagged-freeze phrases · flag grammar · output_format tag) were PRESERVED by carrying them into 3-plan.md — the honest path, not by loosening the tests. Two edits were fragile-regex robustness fixes. - SOFT strategy self-improved: did the book cascade INLINE (one atomic git-mv + path-replace pass) rather than the planned parallel worktree agents — mechanical path-replacement is faster/safer atomically (recorded at §5). - SCOPE HONESTY: the human-requested glossary path fix touched a repo-root copy outside the frozen §5 Scope; the gate flagged scope_violation → extended the SOFT Scope line, re-crossed, re-gated (never a silent out-of-scope build). Verify: full suite 3417/3417 OK (exit 0), net +10 tests; 3-tree skill + 4-tree book + 4-twin template parity; phases byte-budget pool shrinks; refute-read EARNED; advisor 3-lens PASS (one disclosed T4 residue: the GLOSSARY "Ground"/"Contract" term rewrite + flow diagram). Reviewed by Tin Dang, 2026-07-13. author: Tin Dang <tindang.ht97@gmail.com> * feat(method): surface the build-strategy plan-of-action in the freeze report (plan-in-report, T1 plan-legibility) First task of the plan-legibility sub-milestone: at the ONE freeze, the human now sees the AI's full plan-of-action (HOW it will build) — not just the frozen contract SHAPE. `report --decide` already dumped §3 verbatim as raw template markup (illegible as a plan); this extracts the §3 Build-strategy fields into a legible, labeled BUILD PLAN block. What shipped (add.py ×3 engine trees): - NEW `_build_plan(raw3)` — extracts the plan-of-action fields in a fixed order (Scope · Strategy/ordered batches · Approach · Persona · Spawn isolation · Known-problem fixes) as [{label, value}]. Single-physical-line capture (each field is authored one per line): NOT `_capture_wrapped`, whose `Word Word:` continuation boundary misses a label like `Known-problem fixes:` and bled one field's value into the next. Skips a placeholder value (leading `<`, or the bare `./src/` Scope default); strips a trailing ` <hint>`. PURE. - `decide_data` gains ONE additive `plan` key, populated only at the front+unfrozen seam (the freeze); [] everywhere else. Every pre-existing key/value is unchanged; still PURE. - `render_decide` appends a `BUILD PLAN (§3 · how the AI will build)` block after the §3 verbatim dump when `plan` is non-empty — the HARD contract and the SOFT plan stay distinct. - the milestone `--decide --json` payload carries `plan: []` so both altitudes share the shape. - report-template.md (×3, byte-identical) documents the block; net-ZERO byte change (same-guide compression) keeps the reference pool under its frozen budget. - ENGINE_MD5 re-aimed → 7e6ebec0 (add_engine untouched, ENGINE_PKG_MD5 stable). Tests (red→green): - test_decide_digest: +8 (front-seam render · additive json key · placeholder-skip · trailing-hint-strip · empty-omit · gate-seam-absent · purity · a NO-FIELD-BLEED regression that a live dogfood surfaced — Spawn isolation must not swallow Known-problem fixes). - the two exact-key-set asserts + the shared `FROZEN_DECIDE_KEYS` constant re-pinned to include `plan` (a deliberate shape change, never a weakened assertion). - report-template byte-pin migrated 9627→9626; a frozen needle ("the flag count") preserved. Engineering notes: - The §1 ⚠ (exact-key-pinned shape) realized as flagged: the full suite surfaced strays beyond the two named asserts (a 3rd shape test, a rollup key-set, byte/pool budgets, a frozen report needle). Recovery each time = step back to tests (tripwire-safe), fix, re-cross. - The Optimization-stance ⚠ (wrapped-list capture) ALSO realized: a live dogfood render proved the bleed was real, not hypothetical — the regression test + single-line rewrite close it. The dogfood is why the block ships clean. Verify: full suite 3426/3426 OK; 3-tree engine + 3-tree skill parity; reference pool under budget; refute-read EARNED; advisor 3-lens PASS (no residue). Gate PASS by Tin Dang 2026-07-13. author: Tin Dang <tindang.ht97@gmail.com> * feat(method): structured task+milestone relations with an advisory guard (relations-surface, T2 plan-legibility) Add two NON-BLOCKING relation types — `extends` and `relates-to` — as siblings to the existing BLOCKING `depends_on`, declared per task and per milestone, surfaced at status, and validated (dangling/self) by check. The wave scheduler stays a pure depends_on DAG: extends/relates_to never enter _edges_fingerprint, so the schedule is unchanged (test-proven). Task altitude: - new-task --extends / --relates-to seed migration-tolerant state keys (old tasks with no key read []), reusing _parse_deps. - _task_relations reader (fail-safe []-default), status task-row ext=/rel= segments, and a `relations: N dangling · M self` status line. Milestone altitude: - _milestone_relations parses the MILESTONE.md HEADER (pre-first-`##` region, so a per-task depends-on: row is never mistaken for a milestone edge); fail-safe on a missing/garbled/old doc. Advisory guard: - _relations_health is PURE over (state + milestone docs) -> findings {dangling, self_relation}; an archived (PASS-done) target resolves, mirroring depends_on. cmd_check resolves + flags; the guard never writes, never blocks a gate. Vocabulary + where-declared: - GLOSSARY.md.tmpl gains depends-on / extends / relates-to terms; TASK.md.tmpl (autonomy header) + MILESTONE.md.tmpl (relations: header line) name where each relation is declared. test_relations.py: 15 red/green tests (state records, migration-tolerant reader, status show/silent, check dangling/self, health archived-resolves/writes-nothing, milestone header parse/migration/not-confused-with-task-rows, edge-fingerprint invariance, glossary vocab, template naming). Full suite 3441 green. Engine housekeeping: ENGINE_MD5 re-pinned c9c2a671 -> 9311ec35 (relations code + a docstring altitude->level slang fix); SEAMS.md _declared_scope anchor re-pinned 5402 -> 5491; TASK.md.tmpl net-compressed sibling header comments to absorb the relations note under the frozen 12400B/2850B lean ceilings; canonical + bundled engine trees synced byte-identical. Scope note: the frozen §5 bare tokens mis-resolved under the task dir; corrected to the two walked engine-tree directory tokens and re-baselined via a human-signed re-cross (add.py re-cross --by "Tin Dang") — §3 stayed frozen, no test touched. author: Tin Dang <tindang.ht97@gmail.com> * chore(method): close plan-legibility milestone — ship review + exit criteria (2/2 PASS) Both tasks done and committed: - plan-in-report (331306b) — freeze report surfaces the FULL §3 build-strategy plan-of-action (the human approves HOW, not just WHAT). - relations-surface (ed48dc5) — structured depends-on·extends·relates-to at task + milestone altitude, status-surfaced, advisory dangling/self guard. Fill the Close ship-review (ship-by-domain · cross-task evidence · goal-met map), check all three exit criteria, define release steps (defer the package cut — this is a method/engine change to bundle in a later release scope). milestone-done recorded status=done; RETRO.md generated. Full suite 3441 green. author: Tin Dang <tindang.ht97@gmail.com> * docs(method): align the book GLOSSARY + diagram source to the plan phase (book-plan-align, T4 expectations-first) Close the book-narrative debt guides-and-skill (T3) deferred: the book now names the expectations-first flow the engine actually runs. - appendix-c-glossary.md (×4 trees): ADD a **Plan** term (the unified step 3 — ground → freeze the contract → build strategy, exactly one human approval); REWRITE **Ground** as Plan's FIRST part (drop "phase-0 preamble / step 0 / the phase before Specify"); **Contract** gains a "frozen within the Plan phase" clause (meaning intact); the Grounding-map term's stale "§0 GROUND" → "the GROUND part of §3 PLAN". - appendix-g-references.md (×4): "ADD's specify and contract phases" → "specify and plan phases". - add-method/diagrams/prompt-flow.txt: the flow diagram's image-gen SOURCE card 3 "Contract / freeze the shape" → "Plan / ground the code, freeze the contract" (matches CHECKLIST.md + the ch02 mermaid). The add-flow.png RASTER regeneration is an image-model + human visual-gate step (flagged in §7, not producible headless) — the tracked source is now correct. - test_ground_prose.py: new PlanPhaseGlossaryTest pins the **Plan** term, the redefined **Ground**, and a stale-prose grep-guard (forbids phase-0-preamble / as-step-0 / phase-before-specify / the-contract-phase / the-ground-phase across the book; exempts a "formerly …" history bridge; non-vacuous — still fires on a live stale claim). EngineUntouchedByBookAlignTest pins add.py == ENGINE_MD5. Docs-only: engine byte-identical (ENGINE_MD5/PKG unchanged). Full suite 3446 green. Freeze approved by Tin Dang (v1); gate auto-PASS under autonomy:auto. author: Tin Dang <tindang.ht97@gmail.com> * chore(method): close expectations-first milestone (4/4 tasks, 8/8 criteria) Close the expectations-first sub-milestone: the lifecycle now collapses the former `ground` + `contract` phases into ONE `plan` phase carrying the single human freeze, so the flow runs 7 work phases with one approval instead of 8 with grounding scattered per task. Shipped across three domains: - tooling : PHASES reordered specify→scenarios→plan→tests→build→verify→ observe→done; freeze moved to plan→tests; §3 PLAN template renders Grounding/Contract/Build-strategy sub-blocks; legacy ground/contract task states migrate; MILESTONE.md gains `## Ground`. - skill : phases/3-plan.md unifies the retired 0-ground + 3-contract guides; SKILL.md + sibling guides read the 7-work-phase order. - book : ch02 flow + mermaid + CHECKLIST aligned; GLOSSARY defines Plan + redefines Ground as Plan's first part + a Contract "within Plan" nod; appendix-g + diagram render-source read "Plan"; a stale-prose grep-guard pins it. add-flow.png raster regen = human visual gate. Evidence: 4 tasks each gate=PASS (plan-phase-core, milestone-ground-seed, guides-and-skill, book-plan-align); full suite 3446 green; ENGINE_MD5 9311ec35 / ENGINE_PKG_MD5 28212a55 re-pinned; ×3/×4 tree byte-parity holds. RETRO.md written. author: Tin Dang <tindang.ht97@gmail.com> * fix(method): sync dormant task marker missed by the plan-phase migration The expectations-first state migration moved cross-platform-inode-reuse-stress ground->specify in state.json, but the dormant task file marker only syncs on touch — add.py check flagged marker=ground vs state=specify. Sync the marker + its phase-order comment to the 7-work-phase flow. check: 730 passed, 0 failed. author: Tin Dang <tindang.ht97@gmail.com> * fix(method): normalize pilot-cwd-hardening's freeze stamp to the audit grammar The seam-audit CI job (and local `add.py audit`) exits 1 on `unstamped_freeze pilot-cwd-hardening`: the task WAS frozen and gated PASS (add-bench milestone, merged in PR #142), but its stamp cited the approval precedent prose-first instead of the canonical `approved by <name>` form the audit regex requires. Reword the same recorded approval into grammar — no new approval invented, the precedent citation kept in the parenthetical. audit: exit 1 -> 0 locally; this was one of the two CI failures pre-existing on main (the other is the known lock-reclaim TOCTOU flake in test_project_scope_lock, re-run by this push). author: Tin Dang <tindang.ht97@gmail.com> --------- Co-authored-by: Tin Dang <tindang.ht97@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
quality-floors — 4 zero-turn input-dialect floors from the WV1 wm2 miss
Converts the WV1 benchmark's wm2 root cause — the lean arm's own red suite spoke naive timestamps, a friendlier dialect than the spec's own
Z-suffixed examples, so an aware/naive datetime crash shipped green at 0.80 fidelity while full-ceremony add-main caught it — into narrow, ~zero-turn mechanical floors. The lean lane keeps its −27–33% cost; the failure class is closed without re-added ceremony.The 4 floors (3 tasks, each gate PASS)
aware-iso-timestampvalue no declared §4 test file speaks;add.py checklists gapped active tasks underdialect_gapf9d2303Boundary:line;freezerefuses a placeholder/empty value (boundary_unfilled, both freeze paths, validate-then-write, absent line grandfathered)b2483be### Deep checks— audit-counted by the existingshallowmachinery, zero engine code — plus a condensed fast-template twinbb72452sensitivity.md×3: datetime/money/timezone arithmetic ⇒ declaredata(full lane), wm2 evidence cited; prose, no new sensitivity tokenbb72452Balance principle (human-agreed): every floor is a mechanical check at a seam that already existed — the only new refusal (floor 2) fires at a gate the fast lane already had.
Engine / pins
ENGINE_MD5→16cd7cca(floor 2cmd_freezeguard) ·ENGINE_PKG_MD5→710a009f(floor 1_DIALECT_CLASSESregistry inconstants.py)scope-token-grammarpin re-aimed x18 →add.py:5337sensitivity.mdproseTests
test_spec_dialect_floor.py(6),test_dialect_vocab_lines.py(6),test_fast_boundary_line.py(9)test_fast_lane_skipsgained a shared_fill_boundaryhelper; the fast-lane empty-git-diff guard was re-pinned from a task-local check to its durable invariant (skip machinery stays fast-lane-only)add.py check: 715 passed, 0 failedOpen follow-up (not fixed here)
One
[SPEC · open]delta recorded at floor 3's gate: floor 1's_dialect_gapsscans the §3 raw body, which includes the engine's ownFreeze mode: …<ISO timestamp>stamp — so this very task's crossing warnedaware-iso-timestampon its own freeze stamp (a self-referential false positive found while dogfooding). Candidate fix: scan only §3 fenced blocks, or strip the Status/Freeze-mode lines. Left as a future task.Known CI non-blockers (pre-existing, not introduced here)
contract_report_unrecordedon 3 older risk-proportional tasks — deliberately not fabricated toReported: yestest_project_scope_locklock-reclaim TOCTOU flake (py3.10) — known-waived, passes on re-run