Reverses the BDR-066 rejected alternative (opus pins on audit agents):
session default is now Fable, so inherit burned Fable quota on every
dispatched audit/challenge. analyzer, plan-challenger, seo/geo/
validator-analyzer pinned model: opus; onboard's 6 general-purpose
audit dispatches carry model="opus"; tour Phase B repointed.
interviewer + client-handover-writer stay unpinned (inline-load only,
a pin there is inert). settings.json default: claude-fable-5[1m].
Census flipped: model-routing §3 + new §11 (61 pass), loops-light 35,
full make test green.
After a plan/reflection is elaborated and before it executes, three fresh blind
sub-agents (correctness / robustness / simplicity) attack it on the big model;
the main loop RE-THINKS every aspect a BLOCKER lands (a named plan change, or
[deferred]) and re-challenges once if the plan materially changed. Advisory into
each skill's existing human gate — the human stays the decider.
- lib/challenge-plan.md — reusable phase: fail-safe (never fail open),
severity-driven (any single-lens BLOCKER = must-address), RE-THINK loop
- agents/plan-challenger.md — challenger role (read-only, big-model per BDR-066)
- lib/tests/plan-challenger.test.sh — 41-assertion structure lock
- wired into 11 orchestrators: ship-feature/init-project/feat/bugfix (build-plan),
onboard/audit-delta/code-clean (proposals), seo/geo/harden/web-validate (fix-bundle)
Hardened by dogfooding: 3 blind challengers reviewed this feature's own v1 plan
and caught 4 BLOCKERs (fail-open, consensus-buries-lone-finding, wrong model
tier vs BDR-066, false on-disk-plan premise) — all fixed here.
code-cleaner is now a pure fix executor (was audit+gate+execute). Reroute the two
read-only-audit consumers (onboard STEP 6, tour Phase B) to a big-model agent
(general-purpose/analyzer) — an audit must stay on the big model, never the sonnet
executor. Refactor now runs on sonnet inside the executor (inline-load pin was inert).
- A2: graphify build flag --output -> --out
- A4: ROADMAP xref points to the real /onboard add gsd path, not a nonexistent STEP 9 decision
- A5: exact gitflow init commit message (matches lib/gitflow.sh:163)
- A6: bare skill names (design-review, browse) — no gstack: namespace exists
- A7: eval pattern for recommend_anim_install_cmd (the function only echoes; must eval its output)
ship-feature: STEP 0e CONTRACT (request verbatim) → ENRICHED at the STEP 3
validation gate (design criteria appended [gated <date>], the human
micro-gate) → STEP 5 VERIFY+SECURE judges the branch against the ENRICHED
contract via the shared include. Distinct axis from STEP 6 code review, both
run (LRN-095).
init-project: contract seeded from the PROJECT BRIEF (V1 features → criteria)
→ ENRICHED at VALIDATION GATE #1 → STEP 9 VERIFY+SECURE. Adds the security
gate init-project previously lacked (was deferred to a later /onboard).
onboard: explicit NO verify-loop — it produces an audit report, not a change
to verify against a request; contract is scope-only, security-auditor runs
MODE audit (report-only), never a gate. Documented to prevent a misplaced
symmetry loop (BDR-050: dev pipeline != audit).
lib/tests/no-vacuous-locks.test.sh: deterministic backstop for LRN-093 (2nd
recurrence in this chantier → the advisory alone did not hold). Refuses a
literal \n in any grep/tf/tr_/tn pattern across lib/tests/*.test.sh;
flip-tested against a synthetic offender so the guard proves it bites.
lib/tests/loops-heavy.test.sh: 18 structure locks green.
Behavioral dogfood (both vigilance points, real): (1) enrichment — a fresh
verifier reads and checks a [gated] design criterion (ECARTS naming it
precisely); (2) escalation — 3 consecutive ECARTS on the same criterion →
orchestrator STOPs at the max-3 bound + presents the CONTRACT-vs-REALIZED
table, no 4th loop, no commit. First real exercise of the infinite-loop guard.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XpphkdTosUzokBDNG7PToS
agents/security-auditor.md: fresh read-only-on-code SAST gate. Pinned
rulesets p/security-audit + p/secrets + p/owasp-top-ten (owasp REQUIRED —
measured: the 2-ruleset baseline misses SQLi + path-traversal entirely on
realistic Flask code), never --config auto, never auto login (BDR-048).
Severity map: secrets ERROR → CRITICAL, other ERROR → HIGH (block),
WARNING/INFO → reported. gate mode (diff, no Write) vs audit mode (Write
only to REPORT, rule-locked). DEGRADED (semgrep absent) still runs the
checklist and still blocks — never a vacuous pass (LRN-048). Anti-gaming:
a new un-gated nosemgrep suppression is BLOCKING. PROOF mandatory, mute
auditor never a PASS, blind (no iteration history), blocks HIGH/CRITICAL
only (LRN-047).
Grafts: onboard STEP 6 L3a dispatches it in audit mode (report
.onboard-audit/semgrep.md) in BOTH gstack branches — complement to cso
(cso is a gstack submodule, unmodifiable); synthesis picks it up via the
existing .onboard-audit/ sweep. audit-delta security axis runs the SAST
pass first, folds findings into the existing gate/fix/re-verify flow.
lib/tests/security-auditor.test.sh: 28 structure locks green, shellcheck
clean. Behavioral dogfood (fresh agents on a planted fixture):
BLOCK(9) on the vuln commit (2 secrets→CRITICAL, semgrep+checklist
complementarity — checklist caught the 6 semgrep missed off-context);
BLOCK(1) on a new nosemgrep suppression (understood semgrep's 0 was the
mask); DEGRADED → BLOCK(7) on grep-detectable secrets with semgrep hidden.
FP measured on real repos (faunosteo, game): owasp adds only hygiene
findings, contained by diff-scoping.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XpphkdTosUzokBDNG7PToS
Strip the disable-model-invocation frontmatter key from all 19 editable skills. Absent = default = model invocation enabled. 8 were 'true' and blocked the model AND orchestrators from self-routing (status, plugin-check, analyze, onboard, refactor, init-project, pdf-translate, ship-feature) — contradicting the CLAUDE.md skill-routing rules. The other 11 were 'false', a no-op noise line.
The setting is binary (no per-caller granularity), so enabling orchestrator chaining also enables model auto-fire — accepted. Genuinely destructive operations remain guarded by the careful/guard hooks, independent of this flag.
Capitalized: BDR-019 (decision), LRN-026 (learning), journal 2026-06-09.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two issues fixed across personal skills:
1. Five skills had frontmatter > 1024 chars (spec limit per agentskills.io):
- client-handover: 1920 → 924
- doc: 1390 → 734
- seo: 1378 → 885
- geo: 1189 → 734
- validate: 1050 → 742
All now compliant.
2. Three orchestrators had workflow-summary descriptions that create
"shortcut" risk per writing-skills CSO (Claude may follow the
description instead of reading the full skill body, skipping steps):
- ship-feature: "design → plan → implement (TDD) → review → finish"
→ "Use when shipping a feature end-to-end — needs… (9-step pipeline)"
- init-project: "interview → design → scaffold → implement (TDD)"
→ "Use when initializing a brand-new project from scratch — needs…"
- onboard: "detect archetype, install claude-config, run full audit…"
→ "Use when bringing an existing repo into the framework — needs…"
Descriptions now follow "Use when [triggers]…" pattern. Workflow detail
preserved in skill body where it belongs. Triggers consolidated and
trimmed of duplicates.
Co-Authored-By: Claude <noreply@anthropic.com>
Adapt the two skills that scaffold project config + the underlying
onboarder agent + the dotfiles-meta archetype folder tree to the new
governance layout. AUDIT_* files go to .claude/audits/, TODO goes to
.claude/tasks/, and .claude/memory/ is seeded from ~/.claude/templates/memory/
with all 5 registries (decisions/learnings/blockers/journal/evals).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
STEP 4.5 extracts the archetype's "Implications" (Surface sécurité) and
"Typical pain points" sections into .onboard-audit/archetype-context.md.
STEP 6 cso dispatch reads it and filters checks per category: web vulns
(XSS/SQLi/CORS/CSP/CSRF/HTTPS) only on framework/api/ecommerce/cms,
embedded-specific checks (buffer overflow, secure boot, JTAG, OTA sig)
only on embedded; library/cli/infra/data-science/desktop each get their
own focused section. Previously the fallback prompt searched for web
vulnerabilities even on firmware projects.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Move discovery, interview, archetype detection, audit pipeline, and
validation gates from the onboarder agent into the /onboard skill as
a 9-STEP orchestrator (STEP 0 plugin-check → STEP 9 sequenced backlog).
The onboarder agent becomes a pure config generator: takes a prepared
brief, writes CLAUDE.md / settings.json / .claudeignore / tasks/ scaffold.
No more interview or filesystem scanning in the agent.
Agent shrinks 263 → 86 lines; skill grows 15 → 847 lines.
Co-Authored-By: Claude <noreply@anthropic.com>
Bastien Chanot