Skip to content

CHANGELOG.md

Latest commit

 

History

History
339 lines (249 loc) · 45.3 KB

CHANGELOG.md

File metadata and controls

339 lines (249 loc) · 45.3 KB
Raw

Changelog

All notable changes to the Automotive Skills Suite. Maintained by the autonomous daily standup. Entries are grouped by intent (feat / fix / polish / docs) and move from [Unreleased] into a dated section at each weekly release.

[Unreleased]

W40 (2026-09-28 → 2026-10-02). Accumulating since v2026.09.W39 (2026-09-26).

Docs

  • W40 PLAN (Mon 2026-09-28): opened #66–#70, the carried HIGH backlog (sw-arch-builder, sw-sr-builder, sw-fmea-builder TSC/HSI column shifts; safety-case-builder truncated reader; bus-load-analysis-checklist-reviewer truncated CHECKS dict) (4da1631)
  • September monthly KPI report, docs/monthly/2026-09.md (f7ece99)
  • W41 (2026-10-05 → 2026-10-09) DOCS roll (Fri 2026-10-09): no skill, script or doc commits landed this week — the W41 PLAN and Tue–Thu POLISH slots did not commit to the repo, so there are no feat: / fix: / polish: entries to group and no example README stubs owed. #66–#70 remain open and unaddressed; STATUS regenerated via scripts/regen_status.py.
  • W42 (2026-10-12 → 2026-10-16) DOCS roll (Fri 2026-10-16): one commit landed this week — the Tue 2026-10-13 POLISH run. No Mon PLAN, Wed or Thu commits; #66–#68 and #70 remain open and unaddressed.

Fix

  • safety-case-builder — W42 Tue (2026-10-13, #69): multi_xlsx_reader.py was truncated mid-statement (retu), raising NameError on any input. The return of the four-artifact dict was restored and the .skill repackaged; the deferred office/__init__.py file-tree doc fix landed with it. Verified by ast.parse, import + signature check, zip integrity and a cross-check of return keys against every artifacts.get(...) in generate_safety_case.py; the full 15-tab build was not re-run here (96afe77)

Known issues (carried, not fixed this week)

  • #70 bus-load-analysis-checklist-reviewer (CHECKS dict truncated, SyntaxError), #66–#68 (sw-arch-builder, sw-sr-builder, sw-fmea-builder TSC/HSI column shifts). Example input examples/sample_safety_case_input_esc.json still has hardcoded /sessions/... paths.
  • No example README stubs owed: safety-case-builder already has examples/safety-case-builder/README.md.

[v2026.09.W39] — 2026-09-26

W39 (2026-09-21 → 2026-09-25). Accumulating since v2026.09.W38 (2026-09-19). Rolled by the Friday DOCS run (2026-09-25) and shipped by the Saturday RELEASE run (2026-09-26).

Fix

  • cs-goals-builder — W39 Tue (2026-09-22, #62): Cybersecurity_Property was hard-coded Confidentiality on every CSG though the threat's STRIDE category determines it unambiguously. _build_stride_index() added beside the existing _build_asset_index() (same tab, 05_Threat_Scenarios, already indexed by scenario ID) so the property is now derived per row; unmatched rows report Not Determined with a stderr warning instead of silently defaulting to Confidentiality. Verified on six scenarios — the shipped sample, an Information-Disclosure-stripped variant, the 09_Risk_Determination fallback path, a TARA missing 05_Threat_Scenarios, an unrecognised STRIDE value, and a full cell-by-cell before/after diff (4604f67)
  • 24 checklist-reviewer archives (batch) — W39 Wed (2026-09-23, #63): _find_header_row's >= 3 populated-cell gate hid any two-column tab from its reviewer suite-wide. Measurement showed the issue's own proposed blanket >= 2 would have moved two already-correct headers backward (control-plan 02_Header_Block row 11→3, ppap PSW row 17→5), so the gate ships as "prefer ≥3, fall back to 2" instead — proven zero-diff across 2,616 checklist rows built from all 24 paired outputs before/after. Neither of the two symptoms the plan named turned out to move a verdict (cs_concept_probe never reads the affected tab; sw_fmea_probe feeds only an informational count), so the issue was downgraded HIGH → MED in its closing comment. The sweep also found two dead-on-arrival skills — safety-case-builder and bus-load-analysis-checklist-reviewer, both broken since their initial commits — left unfixed this run; see Known issues (50ba884)
  • cs-concept-builder — W39 Thu (2026-09-24, #65): cs_goals_reader.py read the CS Goals tab positionally starting at row 5 against a row-1 header. 3 of 15 goals (firmware injection, calibration tampering, CAN spoofing) were never read, every CSR's CAL echoed the goal's text instead of a rating, and every node inherited CAL-1. Filed MED, measured HIGH. Fixed with a header-spelling lookup table; verified from the repacked archive with all 15 goals present under the right header and 0 verdict moves across cs-concept-checklist-reviewer's 29 checks. This was #64's only known-bad chain edge, watched failing before the fix landed (6d0118f)

Feat

  • Column-level chain-contract audit — W39 Thu (2026-09-24, #64): scripts/column_contract.py (new, 503 lines) extends chain_contract_audit.py from tab-existence checks to column-name and data-start-row assertions per chain edge. Landed by watching the new layer fail on #65's edge first — committed before the fix — then repairing #65 behind it. Beyond #65 it reports 13 further BREAKs, all dynamically confirmed by building the real chain rather than reading source: sw-arch-builder reads 05_TSR_Catalog two columns off, so its "SW" in Type filter tests the FSR_ID and every SW-TSR is dropped (0 imported), and reads 04_Safety_Mechanism_Catalog and the SW-partition table off by one column each; sw-sr-builder reads 06_HSI_Specification one column right of where it should from column 2 on; sw-fmea-builder's tab-name scan looks for partition/sw_tsr, which no TSC tab contains, and imports nothing from the TSC, silently. The entire TSC→software branch has been running on misread or empty input while the tab-level audit reported every one of those edges MATCH (5dcce3d)

Docs

  • W39 weekly plan (Mon 2026-09-21): opened #62–#65 from the prior Sunday's triage handoff; the _find_header_row candidate was measured down from "suite-wide HIGH" to "24 archives carry it, 2 have a live symptom" before being slotted, and the regression-harness item was held back deliberately to keep the week to four targets instead of eight (ea692a7)
  • W39 DOCS roll (Fri 2026-09-25, this section): 18 example README stubs added — apqp-plan, aspice-assessment, control-plan, cs-concept, cs-goals, dfmea, hsi, hw-architecture, hw-safety-reqs, pfmea, ppap, safety-case, safety-plan, sw-arch, sw-fmea, sw-hsis, sw-sr, tara checklist-reviewers — for the 18 of #63's 24-archive batch that had no examples/<skill>/README.md. These are stubs sourced from each skill's own frontmatter description and confirmed builder pairing, not from execution: no real check counts or sample output, flagged as such in each file, and left for a future POLISH pass to deepen. cs-goals-builder and cs-concept-builder already had READMEs, rewritten this week during #62 and #65, and were left alone. STATUS regenerated via scripts/regen_status.py — 76/76 paired (2 via alias registry), 7 🟢 fresh / 69 🟡 stale / 0 orphan (this commit)

Known issues (found this week, deliberately not fixed)

  • Three column-level BREAKs from #64, not yet filed as issues — sw-arch-builder's TSC reader (every SW-TSR dropped; safety-mechanism catalog and SW-partition table both misread), sw-sr-builder's HSI column shift, sw-fmea-builder's TSC tab-name scan matching nothing. All HIGH, all dynamically confirmed by running the real chain; exact column pairs and reproduction steps in docs/chain-contract-audit.md. Should be fixed by header-name lookup, the same shape as #65, not by re-counting positions
  • safety-case-builder — multi_xlsx_reader.py truncated mid-statement (retu), NameError on any input. Rated "exemplary" by a 2026-06-17 polish pass that never ran it. A one-line verified fix is written out in docs/skill-polish-log/reviewer-find-header-row-batch-63.md but not applied — one skill per POLISH slot
  • bus-load-analysis-checklist-reviewer — check_definitions.py's CHECKS dict truncated at entry 18 of 28, SyntaxError on import. The fix is mechanical (all 28 check functions already exist; only the mapping entries are missing) but likewise not applied this run
  • column_contract.py cannot see the #65 edge it just fixed. cs-concept's new header-spelling reader is the right design and a genuine blind spot for a positional-assertion audit; the gap will widen as more readers move to name-based lookup, which is why the reviewer-probe regression harness (carried, HIGH, unfiled) is still needed as the dynamic complement
  • cs-concept does not model CAL-4 at all — generator, verification matrix and SKILL.md line 112 all stop at CAL-3, so 160 CAL-4 CSRs in the sample still inherit CAL-3 (MED, spotted in passing, not filed)
  • cs-goals-builder writes goal text into 02_TARA_Echo.Threat_Description (LOW-MED, spotted in passing, not filed)
  • cs-concept-checklist-reviewer has no check asserting the echoed CS-Goal count equals the upstream count, or that every CSR's CAL falls in CAL-1..4 — a genuine reviewer-finding candidate surfaced by #65's own fix, MED, not filed

[v2026.09.W38] — 2026-09-19

W37 + W38 (2026-09-07 → 2026-09-18). Accumulating since v2026.09.W36 (2026-09-05). Rolled by the Friday DOCS run (2026-09-18) and shipped by the Saturday RELEASE run (2026-09-19). No v2026.09.W37 tag exists and none was backfilled — see RELEASES.md.

Five scheduled runs did not fire across these two weeks. W37 produced only its Monday PLAN and Tuesday POLISH — the Wed/Thu POLISH slots, the Friday DOCS roll and the Saturday RELEASE are all absent from the log, so two of W37's three targets (#60 mbse wiring, #61 test-case-catalog / flexray-config) were opened and never worked. W38 is the mirror image: no docs/weekly/WEEK-2026-W38.md and no commits dated 2026-09-14, so the Monday PLAN never ran, but all three POLISH slots did. Those three selected by rule (c) — least-recently-touched builder — because the open-issue queue is empty (#59–#61 were closed outside the automation). Net effect: no v2026.09.W37 tag exists, this is the first DOCS roll since 2026-09-04, and the week's work is sound but unplanned.

Fix

  • traceability-matrix pair — W37 Tue (2026-09-08, #59): the pair documented as non-functional in W36 was repaired in dependency order. generate_trace.py rewritten so input actually flows into all 11 advertised tabs with coverage and orphans derived rather than emitted as headings; trace_probe.py now locates metrics and orphan rows by label instead of a fixed row window, so orphan_tests populates. TM005/TM006/TM007 corrected, TM008/TM009 implemented, and the 12 checks that cannot be machine-verified labelled AUTO-SUGGEST DRAFT instead of falling through to a hard-coded LC. Builder output now scores 0 FC / 20 LC / 5 NO against an empty workbook's 0 FC / 7 NO / 5 NA — the two spreads differ, which they did not before (302660e)
  • hw-safety-reqs-builder — W38 Tue (2026-09-15): hw_sr_overrides was discarded in its entirety and the run still exited 0. derive_hw_srs_from_tsr matched overrides[tsr_id] exactly, but the shipped examples/sample_hw_sr_input_esc.json keyed its two overrides on a stale three-segment HW-TSR ID (TSR-001-01) while tsc-builder emits four segments (TSR-001-01-01). The generic <TBD: ADC_CHx> template shipped in place of the analyst's text, element and DC% target, with a summary reporting hw_srs_derived: 825 as though refinement had applied. Overrides now resolve, unmatched keys raise a named WARNING on stderr, and the JSON summary carries overrides_supplied / overrides_applied / overrides_unmatched. Same class as #54 — a documented input path that accepts data and drops it (a8f430d)
  • cs-goals-builder — W38 Wed (2026-09-16): the builder could not read the TARA workbook its own upstream produces. tara_reader resolved columns against snake_case names with positional fallbacks; tara-builder emits human-readable headers (Cybersecurity Goal, Risk Value), so no alias matched for asset or cs_goal_text and both took their positional default — every CSG shipped with the goal text in the Asset column and a bare risk integer as its goal text. Plausible-looking and wrong, carried end to end through cs-concept-builder and cs-architecture-builder. Fixed with _HEADER_ALIASES / _normalize_headers() / _col() (name-first resolution, None rather than a guessed column where no alias matches) plus _build_asset_index(), which walks 05_Threat_Scenarios and 04_Asset_Inventory because the upstream CSG tab has no asset column at all. Also hardened against an AttributeError on non-string header cells (a2af2f1)
  • dia-checklist-reviewer — W38 Thu (2026-09-17): three false findings on the pair's own canonical sample, all one kind — the probe assumed column names and layouts dia-builder does not emit, then reported a compliance failure rather than a read failure. DIAA-8 stuck at PC on seven fully-scheduled joint reviews (milestone/participants vs the emitted Review Name/Attendees/Planned Date); DIAA-14 stuck at PC against 6 in-scope / 5 out-of-scope bullets (_find_header_row needs ≥3 populated cells, 03_Item_Scope is a two-column vertical layout); DIAA-4/5 stuck at NO on 10 rows because combined A,R cells — the builder's own documented default — were compared with v.upper() == "A". Fixed via header aliases, a _scan_vertical_scope() fallback that only runs when the header parse finds nothing, and _raci_tokens(). DIAA moved FC 6 / NO 2 / PC 2 → FC 8 / NO 2; the builder itself was found healthy and untouched (9170271)

Docs

  • W37 weekly plan (Mon 2026-09-07): three targets #59, #60, #61 opened. Its pre-slotting audit killed a hypothesis three previous plans had carried — that dashboard.py gating REJECTED on obligation == "Shall" while check_definitions.py uses Must made the verdict dead code suite-wide. Measured across all 76 reviewers: genuinely dead in 3 archives (all mbse), consistent in 38, no Shall literal at all in 11, and 24 where the obligation is carried elsewhere and which remain an open question. The batch pass collapsed into a three-line add on target 2 rather than occupying a slot (f95f4c9)
  • W38 DOCS roll (Fri 2026-09-18): this section; three example READMEs written against the archives — hw-safety-reqs-builder, cs-goals-builder, dia-checklist-reviewer, plus traceability-matrix-checklist-reviewer carried from the skipped W37 roll. Check and tab counts re-derived from source at write time: dia reviewer 14 CR + 14 DIAA + 5 VA = 33 matches SKILL.md, cs-goals 8 tabs matches SKILL.md, hw-safety-reqs 10 tabs matches SKILL.md — no drift found this week, the first clean docs roll since W33. STATUS regenerated via scripts/regen_status.py — 76/76 paired, 8 fresh / 68 stale / 0 orphan (this commit)

Known issues (found this week, deliberately not fixed)

  • _find_header_row requires ≥3 populated cells. Any two-column tab in any builder is invisible to its paired reviewer, which will then report a content gap rather than a read failure. Confirmed on dia-checklist-reviewer; almost certainly not unique to it. Candidate for a suite-wide sweep — the DIA instance lived 4.5 months
  • chain_contract_audit.py is tab-level only. It reported 0 BREAK across all 16 chains for the entire time the tara-builder → cs-goals-builder payload was corrupt, because it asserts a referenced tab exists and never that the columns line up. A column-level assertion is the highest-leverage tooling change currently available and would retro-catch this class suite-wide
  • cs-goals-builder hard-codes Cybersecurity_Property = "Confidentiality" on all 15 CSGs though the STRIDE category determines it. Left alone: threading a new field through a payload schema three downstream skills read is wider than a single POLISH slot
  • examples/sample_dia_esc.json has a real RACI defect — WP11 Safety Plan has two Accountable parties and no Responsible; WP12 Safety Case has A,R for both. Now that the reviewer reads correctly these surface as genuine NOs. Needs a human call: correct the sample, or relax DIAA-4 to allow declared joint accountability
  • cs-concept-builder 02_CS_Goals_Echo header/data column shift (MED) — spotted in passing on 2026-09-16, belongs to another skill, no issue opened because the queue tooling was not exercised this week
  • traceability-matrix-checklist-reviewer.skill ships three scripts/__pycache__/*.pyc files. Harmless at runtime, contaminating in an archive. Recorded not fixed — repacking a .skill is POLISH work. Worth a suite-wide __pycache__ scan
  • No regression harness runs builder output through the paired reviewer's probe. Every defect fixed this week is of exactly that shape and every one survived months. Worth a PLAN target

[v2026.09.W36] — 2026-09-05

W35 + W36 (2026-08-24 → 2026-09-05). Accumulating since v2026.08.W34 (2026-08-22). Shipped by the Saturday RELEASE run (2026-09-05). No v2026.08.W35 tag exists and none was backfilled — see RELEASES.md.

Two weeks in one section. The scheduled run died after 2026-08-27 and missed the W35 Friday DOCS roll and the W35 Saturday RELEASE, so no v2026.08.W35 tag exists and W35's four commits were never written up here. This DOCS run (2026-09-04) rolls both weeks. Grouped by intent, W35 first.

Fix

  • autosar-bsw-config-builder — W35 Tue (2026-08-25, #54): builder no longer silently drops memory_layout and bus_interfaces from its input; the paired reviewer's C005/C006 are reachable again. Builder and reviewer repacked in one commit (84ea8b4)
  • autosar batch (5 archives) — W35 Wed (2026-08-26, #55): canonical 5,782-byte recalc.py restored and the mis-generated ## Skills inventory heading corrected in the remaining five autosar archives. Repo-wide: recalc.py 152/152 at one hash, bad heading 0/152 (656bdfe)
  • sysml reviewers (4 archives) — W35 Thu (2026-08-27, #56): sysml-block-diagram-checklist-reviewer crashed on every input; one-line probe fix applied and the identical defect found and fixed in the other three sysml reviewers. Domain went from zero working reviewers to four (6e8cce6)
  • mbse reviewers (3 archives) — W36 Thu (2026-09-03, #58): all three mbse reviewers crashed at dashboard.py:138 (flat list passed where {tab: [(check, result)]} expected — the #56 class). One-line fix, verified by execution on each reviewer's own builder output. Domain went from zero working reviewers to three; the 2026-09-01 chain-scan's unresolved Stakeholder Needs row settled by execution as not a break (da1082a)

Feat

  • sysml-block-diagram-builder — W36 Wed (2026-09-02, #57): builder now accepts --input <model.json>; the five placeholder blocks are gone from the generator (no --input now yields an empty template with a stderr warning — behaviour change, see Known issues); schema documented in SKILL.md with a brake-by-wire examples/sysml-block-diagram-builder/sample_input.json. Paired reviewer's BDD-12 now validates both connector ends. This is the reference implementation for the W37 sysml batch (72eefda)

Polish

  • traceability-matrix-builder + reviewer — W36 Tue (2026-09-01, #52, first v&v entry): documented, nothing fixed. Builder ignores its input and emits five heading cells; reviewer certifies that empty workbook at 24/25 compliant because 22 of 25 checks are a hard-coded LC fall-through. Full finding in docs/skill-polish-log/traceability-matrix-builder.md; repair queued for W37 in dependency order (5660048)
  • mbse-system-context-builder — W36 Thu (2026-09-03, #58): first pass on the last never-entered domain. Builder populates five tabs correctly from input; seven of twelve tabs are single-cell placeholders and stakeholder_needs / boundary are silently dropped — filed HIGH in docs/skill-polish-log/mbse-system-context-builder.md, not fixed. examples/mbse-system-context-builder/ (README + ACC sample_input.json) added (da1082a)
  • sysml-block-diagram-builder — W35 Thu (2026-08-27, #56): first sysml pass; two HIGH gaps documented and deliberately not repaired (no input path — since fixed under #57 — and reviewer check-count drift) (6e8cce6)

Docs

  • W35 weekly plan (Mon 2026-08-24): three targets #54, #55, #56; #56 opened (6ed7fe5)
  • W36 weekly plan (Mon 2026-08-31): deferred #52 slotted first, #57 and #58 opened (30044b2)
  • Monthly KPI report docs/monthly/2026-08.md (2026-09-01): 29 commits, 17 archives touched, 3 releases; best month on every velocity measure (357da3d)
  • Builder-to-reviewer chain scan (2026-09-01, inside #52): all 76 pairs scanned for reviewer-probed sheet names the paired builder never emits. Confirmed breaks: traceability-matrix, test-case-catalog (probes Test Case Inventory, builder emits Test Cases), flexray-config (probes Title, builder emits Title_Page). This falsifies the premise on which docs/chain-contract-audit.md excludes builder-to-reviewer pairs — #46 should be re-scoped, not closed
  • W36 DOCS roll (Fri 2026-09-04): this section; 4 reviewer-side example stubs added — sysml-block-diagram-checklist-reviewer, mbse-system-context-checklist-reviewer, mbse-model-architecture-checklist-reviewer, mbse-requirements-allocation-checklist-reviewer — written against the archives, which surfaced the check-count drifts below. STATUS regenerated via scripts/regen_status.py — 76/76 paired, 10 fresh / 66 stale / 0 orphan (this commit)

Known issues (found these two weeks, deliberately not fixed)

  • Behaviour change: sysml-block-diagram-builder with no --input emits an empty template, not the old five placeholder blocks. Anything scripting the bare command should point at examples/sysml-block-diagram-builder/sample_input.json
  • Reviewer check-count drift, counted from check_definitions.py on 2026-09-04: sysml-block-diagram-checklist-reviewer advertises 28 checks / 7 tabs, ships 15 / 3; mbse-system-context-checklist-reviewer advertises "25+", ships 6; mbse-model-architecture-checklist-reviewer advertises "30+", ships 8; mbse-requirements-allocation-checklist-reviewer advertises "28+", ships 6. Same class as the W33/W34 entries; awaiting the standing-item-9 decision on whether this task authors checks
  • dashboard.py in the mbse reviewers counts major issues as NO and obligation == "Shall" while the checks use Must/Should, so REJECTED can never fire — a {} workbook scoring 4 NO reports CONDITIONAL APPROVAL. Grep suggests the same split exists elsewhere; repo-wide check queued
  • traceability-matrix pair non-functional (above); test-case-catalog and flexray-config each have a one-name sheet mismatch that may be a one-line fix — unconfirmed

[v2026.08.W34] — 2026-08-22

W34 (2026-08-17 → 2026-08-22). Accumulating since v2026.08.W33 (2026-08-15). Shipped by the Saturday RELEASE run (2026-08-22).

Polish

  • Repo-wide chain-contract audit — W34 target (Tue 2026-08-18, #46) after five consecutive weeks of being carried and never started. Landed only because Monday descoped it to builder-to-builder reads and slotted it first instead of last. scripts/chain_contract_audit.py + docs/chain-contract-audit.md committed; read-only, modifies no .skill file. Found one silent BREAK — fmeda-builder's TSC reader — filed as #53 (5fa7109)
  • fmeda-builder — (Wed 2026-08-19, #53): the break the audit found, fixed the day after it was found. The TSC reader expected sheet names the current tsc-builder no longer emits, so safety-mechanism allocations were silently absent from the FMEDA worksheet — no exception, no warning, the same failure shape as #43. Repo chain breaks now zero (ec8fc8e)
  • autosar-bsw-config-builder — W34 target (Thu 2026-08-20, #51): first pass on the largest never-polished domain. Gained examples/sample_input.json and a real recalc.py (restored byte-identical from autosar-swc-builder, import verified); the ## Skills inventory heading corrected. Two defects found and deliberately not fixed in the same pass — filed as #54 and #55 rather than expanded into a refactor (15d4c8b)

Docs

  • W34 weekly plan published (Mon 2026-08-17) — 3 targets for 3 polish days, down from W33's 4-for-3; #46 descoped and moved to first slot, #51 and #52 opened (547ced0)
  • W34 DOCS roll (Fri 2026-08-21): [Unreleased] filled with the three polish entries above; 2 reviewer-side example stubs added — fmeda-checklist-reviewer and autosar-bsw-config-checklist-reviewer, the paired reviewers of both builders touched this week. Written against the actual archives rather than the SKILL.md prose, which surfaced two more check-count drifts (below). STATUS regenerated via scripts/regen_status.py — 76/76 paired, 16 fresh / 60 stale / 0 orphan (this commit)

Known issues (found this week, deliberately not fixed)

  • fmeda-checklist-reviewer — SKILL.md says "14 + 14 checks" and labels the Confirmation Review tab "14 generic doc-quality checks"; check_definitions.py registers 18 CR entries, so the true total is 36 (18 CR + 14 FMEDAA + 4 VA), not 28. Found during the Friday stub write
  • autosar-bsw-config-checklist-reviewer — SKILL.md advertises "~30 compliance checks" and the check_definitions.py docstring repeats it; the file defines 9 (C001–C009). Off by more than 3×. Should fold into the #55 batch pass, which already opens this SKILL.md
  • Both are the same drift class as cs-architecture-checklist-reviewer (42 → 43) and both require a .skill zip repack, i.e. POLISH-mode work

Not done this week

  • #52 (traceability-matrix-builder, v&v) was Thursday's slot and was displaced by #53. That was the right call — #53 was a live chain break found by Tuesday's own audit, and fixing it the next day is the audit paying for itself — but it means W34 landed 3 of 4 items, and the v&v domain is still never-polished. #52 should take the first polish slot in W35, the way #46 did in W34

Carried forward

  • Four known defects shipped with v2026.08.W33 remain open and all require a .skill zip repack: cdd-checklist-reviewer dangling references/ pointers; cs-architecture-checklist-reviewer check count 42 → 43 and 14 → 15 VA; the two interacting sotif-analysis-checklist-reviewer probe defects; and the 19 over-length sheet names in docs/sheet-name-length-audit.md. See the v2026.08.W33 section of RELEASES.md

[v2026.08.W33] — 2026-08-15

W33 (2026-08-10 → 2026-08-15). Accumulating since v2026.08.W32 (2026-08-08). Shipped by the Saturday RELEASE run (2026-08-15).

Fixed

  • STATUS pairing regression — the Thursday run (08cf363) reported 74 strict-paired, 2 naming-mismatch orphans, contradicting docs/PAIRING_ALIASES.md, which states that any STATUS regeneration MUST honor the alias table. Root cause is the one Monday's plan already flagged (fbcf48e): "the STATUS generator is not version-controlled, so its fixes evaporate each run." Each day re-derived the pairing logic inline, so the alias handling was rebuilt from scratch — and lost. Fixed at the root by committing scripts/regen_status.py, which parses the alias table out of docs/PAIRING_ALIASES.md instead of hard-coding it; a new alias row is now picked up with no code change. Back to 76/76 paired, 0 orphan (this commit)
  • Two builders mis-domained as safety — msa-gage-rr-builder and spc-chart-builder are Measurement System Analysis (Gage R&R) and SPC tooling, i.e. IATF 16949 quality, not ISO 26262 safety. Nine builders match none of the documented domain prefix rules and were being hand-assigned on each ad-hoc run; all nine are now pinned explicitly in regen_status.py so the label is stable, with these two corrected (this commit)

Docs

  • W33 weekly plan published (Mon 2026-08-10) — 4 targets: #48 stub tail, #43 reader rewrite, #46 audit, plus new issue #50 (cdd-builder, first diagnostics target since W30) (fbcf48e)
  • W33 DOCS roll (Fri 2026-08-14): 3 reviewer-side example stubs added for the paired reviewers of skills touched this week — cdd-checklist-reviewer, cs-architecture-checklist-reviewer, sotif-analysis-checklist-reviewer. Written against the actual archives rather than the SKILL.md prose, which surfaced two documentation drifts (below). STATUS regenerated (this commit)

Polish

  • sotif-analysis-builder — W33 target (Tue 2026-08-11, #48): tabs 03/07/09 ran 39/35/32 characters against Excel's hard 31-character limit; openpyxl warns but writes through, so shipped workbooks were malformed. Shortened and verified (12 tabs, 0 over-limit). Second-order fix: the reviewer's first-keyword-hit probe was reading tab 03 (function list) as the performance-limitation table, which the rename resolves. Also restored cdd-builder's promised sample fixture and dropped three aspirational references/ entries. New docs/sheet-name-length-audit.md records 19 over-length names across 10 skills (386191e)
  • cs-architecture-builder — W33 target (Wed 2026-08-12, #43): cs_concept_reader.py rewritten against the tabs cs-concept-builder actually emits, with old names kept as fallbacks; silent empty parse now raises instead of returning nothing. Verified by execution — 0/0/0 before, 2/6/1 after (b9c8aef)
  • cs-concept-builder — (Thu 2026-08-13, #44): re-verified byte-clean; an EOCD-aware rescan found 0 truly corrupt members across all 152 archives, correcting the earlier 13-archive figure, which was scanner false positives on xlsx zip members nested inside the skill bundles (08cf363)

Known issues (found this week, deliberately not fixed)

  • cdd-checklist-reviewer — SKILL.md instructs the user to read references/methodology.md and references/cdd_checks.md and lists both under "Files in this skill"; neither exists in the archive, which contains only SKILL.md + scripts/. Same dangling-reference class as 9b4660e
  • cs-architecture-checklist-reviewer — SKILL.md claims "42 standard checklist requirements" and "14 verification checks"; check_definitions.py defines 15 va_* checks, so the true total is 43. Both the description and the tab table are one short
  • sotif-analysis-checklist-reviewer — the two interacting probe defects from 386191e remain open: the insufficiency keyword never matches (count is permanently 0) and the header row is counted as data
  • All four require repacking a .skill zip, which is POLISH-mode work; queued for next week

[v2026.08.W32] — 2026-08-08

W32 (2026-08-03 → 2026-08-08). Accumulating since v2026.08.W31 (2026-08-01). Shipped by the Saturday RELEASE run (2026-08-08).

Polish

  • aspice-process-evidence-builder — W32 target (Tue 2026-08-04, #45): performance_records, quantitative_metrics and improvements input keys now render into tabs 06/07/08; the hard-coded metric list survives only as a named empty-input fallback, and the status JSON gained per-section counts so silently dropped input shows up in smoke tests. Closes the remaining third of #45 (f00f1b2)
  • a2l-builder — W32 target (Wed 2026-08-05, #47): normalize_members() added so comma-string members no longer report a character count in Record Layouts / Group Hierarchy; conversion_methods.description documented in the docstring schema (the worksheet already read it). Both input shapes smoke-tested, 13/13 sheets (6fab2ce)
  • safety-program-risk-register-builder — W32 target (Thu 2026-08-06, #49): first pass on the program-mgmt domain — SKILL.md corrected from a claimed 9 tabs to the true 8, three phantom files dropped from "Files in this skill", the dangling references/risk_categories.md pointer removed from the References tab, and a Known-limitations block added for the Risk Trend / Escalation Log empty scaffolds (9b4660e)

Docs

  • W32 weekly plan published (Mon 2026-08-03) — no new issues opened; the week carries the five still-open targets #43, #45, #46, #47, #49 (fcc4c0b)
  • W32 DOCS roll (Fri 2026-08-07): [Unreleased] updated with the three W32 polish entries; 3 example README stubs added — safety-program-risk-register-builder plus the paired reviewers a2l-checklist-reviewer and aspice-process-evidence-checklist-reviewer; STATUS regenerated (this commit)

Release (this snapshot commit)

  • STATUS.md regenerated (76/76 paired, 12 fresh / 64 stale / 0 orphan)
  • RELEASES.md appended with the v2026.08.W32 section
  • CHANGELOG [Unreleased] rolled into this dated section
  • docs/AUTONOMOUS_LOG.md updated with the RELEASE-mode entry

[v2026.08.W31] — 2026-08-01

W31 (2026-07-27 → 2026-08-01). Accumulating since v2026.07.W30 (2026-07-25). Shipped by the Saturday RELEASE run (2026-08-01).

Polish

  • aspice-gap-analysis-builder — W31 target (Tue 2026-07-28, #45): the required-but-never-read assessment.xlsx CLI arg removed and the docs made honest about what the generator actually consumes (67c9072)
  • aspice-improvement-plan-builder — W31 target (Thu 2026-07-30, #45): same ignored-arg class — the unread gap_analysis.xlsx arg removed, docs corrected to match behaviour (299cdad)

Docs

  • W31 weekly plan published (Mon 2026-07-27) — targets: cs-concept→cs-architecture reader rewrite (#43, carried), ASPICE ignored-arg/scaffold wiring (#45, carried), repo-wide chain-contract audit (#46, carried), sotif-analysis first pass (#48, new), safety-program-risk-register first pass (#49, new) (57238bf)
  • W31 DOCS roll (Fri 2026-07-31): [Unreleased] updated with W31 entries; 2 reviewer-side example stubs added for the paired reviewers of skills touched this week; STATUS regenerated (e749617)

Release (this snapshot commit)

  • STATUS.md regenerated (76/76 paired, 10 fresh / 66 stale / 0 orphan)
  • RELEASES.md appended with the v2026.08.W31 section
  • CHANGELOG [Unreleased] rolled into this dated section
  • docs/AUTONOMOUS_LOG.md updated with the RELEASE-mode entry

[v2026.07.W30] — 2026-07-25

W30 (2026-07-20 → 2026-07-25). Accumulating since v2026.07.W29 (2026-07-18). Shipped by the Saturday RELEASE run (2026-07-25).

Polish

  • cs-concept-builder — W30 target (Tue 2026-07-21, #44): 3,361 trailing NUL bytes stripped from the generator; same pass surfaced 13 more corrupt archives repo-wide (691f614)
  • batch NUL-strip, 13 archives — (Wed 2026-07-22): the 13 corrupt archives found Tuesday cleaned in one batch; 9 generators compile again; xlsx false-positive findings corrected (8a855cf)
  • a2l-builder — W30 target (Thu 2026-07-23, #47): axis-points crash fixed, trigger description strengthened, 13-tab output verified (661c4bb)

Docs

  • W30 weekly plan published (Mon 2026-07-20) — targets: cs-concept→cs-architecture chain repair (#43), cs-concept NUL-strip (#44), ASPICE bundle wiring (#45), repo-wide chain-contract audit (#46), a2l polish (#47) (a0c3d91)
  • W30 DOCS roll (Fri 2026-07-24): [Unreleased] updated with W30 entries; 11 example README stubs added for skills touched this week — first reviewer-side stubs, per the touched-this-week rule; STATUS regenerated (this commit)

Release (this snapshot commit)

  • STATUS.md regenerated (76/76 paired, 8 fresh / 68 stale; aspice-assessment-builder aged past the 30-day line)
  • RELEASES.md appended with the v2026.07.W30 section
  • CHANGELOG [Unreleased] rolled into this dated section
  • docs/AUTONOMOUS_LOG.md updated with the RELEASE-mode entry

[v2026.07.W29] — 2026-07-18

W28–W29 (from Mon 2026-07-06). Accumulating since v2026.07.W27 (2026-07-04). The W28 Fri DOCS (2026-07-10) and Sat RELEASE (2026-07-11) runs did not fire, so this section spans two weeks; W28 entries were back-filled by the W29 Fri DOCS roll (2026-07-17). Shipped by the Saturday RELEASE run (2026-07-18).

Polish

  • ppap-package-builder — W28 weekly-target pass (Tue 2026-07-07, #39); embedded fix: over-limit sheet title 05_Element3_Engineering_Approval (32 chars) renamed to 05_Element3_Eng_Approval; two chain-break findings logged (tab-09 name >31 chars vs reviewer probe; Element-18 status-tab probe mismatch) (90288f8)
  • item-definition-builder — W28 weekly-target pass (Wed 2026-07-08, #38); resolved via the alias-documentation path — docs/PAIRING_ALIASES.md created as the canonical builder↔reviewer alias registry; smoke test green, no .skill edits (26bb385)
  • aspice-improvement-plan-builder — W29 polish pass (Wed 2026-07-15); smoke test green (11 tabs), findings logged: required gap_analysis.xlsx CLI arg never read (ignored-arg chain-break, same class as aspice-gap-analysis 2026-07-01) and tabs 04–09 are header-only scaffolds; no .skill edits (26ee19a)
  • aspice-process-evidence-builder — W29 polish pass (Wed 2026-07-15); clean data point for the ASPICE ignored-arg sweep (input JSON genuinely consumed, tabs 00–05 populated); scaffold-only tabs 06–08 finding logged; no .skill edits (696e912)
  • cs-architecture-builder — W29 polish pass (Thu 2026-07-16); smoke test green (12 tabs, 14 CSR allocations), forward chain to reviewer exact (10/10 tab names), but high-severity silent upstream chain-break found: cs-concept-builder output tabs/headers don't match this builder's cs_concept_reader.py (a real concept workbook reads back zero CSRs); coordinated fix deferred, no .skill edits (8304e58)

Docs

  • W28 weekly plan published (Mon 2026-07-06) — targets: item-def pairing (#38), ppap pairing (#39), a2l (#40), sotif (#41), risk-register (#42) (a431b86)
  • docs/PAIRING_ALIASES.md — canonical alias registry for the two name-mismatched pairs: item-definition-builder ↔ item-def-checklist-reviewer, ppap-package-builder ↔ ppap-checklist-reviewer (26bb385)
  • W29 DOCS roll (Fri 2026-07-17): [Unreleased] back-filled with W28 entries (missed 2026-07-10 DOCS run) and updated with W29 polish entries; example README stubs added for ppap-package-builder, aspice-improvement-plan-builder, aspice-process-evidence-builder, and cs-architecture-builder; STATUS regenerated (this commit)

Release (this snapshot commit)

  • STATUS.md regenerated (76/76 paired, 4 fresh / 72 stale; dia/fmeda/fsc/hsi/item-definition reclassified other → safety)
  • RELEASES.md appended with the v2026.07.W29 section
  • CHANGELOG [Unreleased] rolled into this dated section
  • docs/AUTONOMOUS_LOG.md updated with the RELEASE-mode entry

[v2026.07.W27] — 2026-07-04

W27 (Mon 2026-06-29 → Sat 2026-07-04, ISO week 27). Shipped by the Saturday RELEASE run.

Polish

  • apqp-plan-builder — W27 polish pass (Tue 2026-06-30); smoke-tested, polish-log entry appended, STATUS regenerated; no .skill edits (2df3374)
  • item-definition-builder — W27 polish pass (Wed 2026-07-01); reviewed, naming-mismatch finding logged to skill-polish-log, STATUS regenerated; no .skill edits (01f38c5)
  • aspice-gap-analysis-builder — W27 polish pass (Thu 2026-07-02); smoke-tested, ignored-argument finding logged to skill-polish-log, STATUS regenerated; no .skill edits (260c03f)

Docs

  • W27 weekly plan published (Mon 2026-06-29) — targets: odx, autosar-bsw-config, mbse-context, sysml-state, traceability (8847b29)
  • June 2026 monthly KPI report published (Wed 2026-07-01) (35be610)
  • W27 DOCS roll (Fri 2026-07-03): [Unreleased] updated with W27 polish + docs entries; example README stubs added for apqp-plan-builder, item-definition-builder, and aspice-gap-analysis-builder; STATUS regenerated (this commit)

Release (this snapshot commit)

  • STATUS.md regenerated (76/76 paired, 3 fresh / 73 stale; generated-on date advanced to 2026-07-04)
  • RELEASES.md appended with the v2026.07.W27 section
  • CHANGELOG [Unreleased] rolled into this dated section
  • docs/AUTONOMOUS_LOG.md updated with the RELEASE-mode entry

[v2026.06.W26] — 2026-06-27

W26 (Mon 2026-06-22 → Sat 2026-06-27, ISO week 26). Shipped by the Saturday RELEASE run.

Polish

  • 5-why-builder — W26 polish pass (Tue 2026-06-23); example README stub added and polish-log entry appended; no .skill edits (107d490)
  • communication-matrix-builder — W26 polish pass (Wed 2026-06-24); trigger phrasing broadened in description, polish-log entry added, STATUS regenerated (9389a13)
  • aspice-assessment-builder — W26 polish pass (Thu 2026-06-25); embedded fix: capability-level rating rule corrected to ISO/IEC 33020; polish-log entry added, STATUS regenerated (fd66c47)

Docs

  • W26 weekly plan published (Mon 2026-06-22) — targets: 5-why, aspice-assessment, cs-concept, dia, communication-matrix (e2162f8)
  • W26 DOCS roll (Fri 2026-06-26): [Unreleased] updated with W26 polish entries; example README stubs added for aspice-assessment-builder and communication-matrix-builder; STATUS regenerated (57dd9ba)

Release (this snapshot commit)

  • STATUS.md regenerated (76/76 paired, 3 fresh / 73 stale; generated-on date advanced to 2026-06-27)
  • RELEASES.md appended with the v2026.06.W26 section
  • CHANGELOG [Unreleased] rolled into this dated section
  • docs/AUTONOMOUS_LOG.md updated with the RELEASE-mode entry

[v2026.06.W25] — 2026-06-20

Consolidates the unreleased W24 + W25 work; the W24 Saturday release (2026-06-13) did not fire, so these entries ship under the W25 tag. Commit messages carry "W24" labels (one-week drift vs. ISO week 25).

Polish

  • sotif-analysis-builder — W24 #20 review pass (Tue 2026-06-09); polish-log entry added; no .skill edits (a6df28d)
  • safety-case-builder — W24 #21 review pass (Wed 2026-06-17); polish-log entry added; no .skill edits (2bd0c19)
  • control-plan-builder — polish pass (Thu 2026-06-18); embedded fix: sample-count in JSON example corrected 18 → 15 to match the 15 controlled characteristics; polish-log entry added (54e2559)

Docs

  • W24 weekly plan published — targets: classifier extract, sotif, safety-case, control-plan, comm-matrix (2623e14)
  • W24 triage pass — label refresh, #12 typed, #17 delta confirmed, STATUS regen (d8b19ef)
  • W25 DOCS roll (Fri 2026-06-19): [Unreleased] backfilled with W24 entries; example README stubs added for control-plan-builder and safety-case-builder; STATUS regenerated (this commit)

[v2026.06.W23] — 2026-06-06

Polish

  • cs-concept-builder — W23 #1 polish pass (Tue 2026-06-02); W20 findings re-confirmed against unchanged archive, polish-log appended; no .skill edits (description rewrite outside autonomous allowlist) (5b4b006)
  • tara-builder — W23 #2 polish pass (Wed 2026-06-03); new polish-log entry with 1 medium-severity finding (Auto-rating Heuristics internal contradiction) + 3 low-severity items; no .skill edits (22d6409)
  • fmeda-builder — W23 #3 polish pass (Thu 2026-06-04); new polish-log entry with 2 medium-severity findings (Classification ladder unreachable branch; SMvDU non-standard acronym) + 3 low-severity items including 100× unit-convention suspicion in JSON example; no .skill edits (d6afa26)

Docs

  • W23 weekly plan published — targets: cs-concept, aspice-assessment, classify_skill.py extraction (#10), fmeda, tara; carryovers #4 and #5 referenced in place, fresh issues #15 and #16 opened (3af1f6b)
  • W23 example README stubs added for skills touched this week (cs-concept-builder, tara-builder, fmeda-builder) (this commit)
  • W23 CHANGELOG roll: 3 polish entries + 3 docs entries staged under [Unreleased] (this commit)
  • May 2026 monthly KPI report published — 23 commits, 24 distinct skills touched, 3 weekly releases, 100% paired ratio, 7.9% example coverage; SOTIF domain flagged zero-touch in May (f8e940f)
  • v2026.06.W23 weekly snapshot tagged; RELEASES.md updated with W23 section; STATUS regenerated

[v2026.05.W22] — 2026-05-30

Polish

  • uds-services-builder — W22 #1 polish pass; small fixes applied in-skill, findings logged to skill-polish-log, STATUS regenerated (9850780)
  • dfmea-builder — W22 #5 polish pass; findings logged to skill-polish-log, STATUS regenerated (04482aa)
  • hara-builder — W22 #2 polish pass; findings logged to skill-polish-log, STATUS regenerated (ef78172)

Docs

  • W22 weekly plan published — targets: uds-services, hara, cs-concept, aspice-assessment, dfmea; carryover items from W20/W21 absorbed and 1 new tracking issue (#12) opened (973c075)
  • W22 example README stubs added for skills touched this week (uds-services-builder, hara-builder, dfmea-builder) (f9e63f9)
  • STATUS.md regen aliasing applied for item-definition ↔ item-def-checklist-reviewer and ppap-package ↔ ppap-checklist-reviewer so suite stays at 100% paired (f9e63f9)
  • v2026.05.W22 weekly snapshot tagged; RELEASES.md updated with W22 section

[v2026.05.W21] — 2026-05-23

Polish

  • autosar-swc-builder — W21 #8 polish pass; description fix applied, STATUS regenerated (387bbcd)
  • dbc-builder — W21 #7 polish pass; findings logged to skill-polish-log, STATUS regenerated (76b2fff)
  • 8d-problem-solving-builder — W21 #6 polish pass; findings logged to skill-polish-log, STATUS regenerated (e8ff3b2)

Docs

  • W21 weekly plan published — targets: 8d carryover, dbc, autosar-swc, uds-services, classifier-freeze (3320c23)
  • CHANGELOG.md introduced; examples/<skill>/README.md stubs added for skills touched in W21 (1a5ca80)
  • v2026.05.W21 weekly snapshot tagged; RELEASES.md updated

[v2026.05.W20] — 2026-05-16

Polish

  • hara-builder, cs-concept-builder, aspice-assessment-builder — W20 polish passes; findings logged to skill-polish-log

Docs

  • W20 weekly plan published; RELEASES.md created; first autonomous-cadence weekly snapshot tagged