Layer C of the plan written after the 2026-09-21 wipe (BDR-095): a reviewer
sub-agent traced `lftp mirror --delete` against a local file:// tree, the
prose tiers named neither lftp nor a local trace, the brief had authorized
it, and four days of commits had never left the machine.
- gitflow: `start` pushes the branch with its upstream, merge targets are
pushed after each merge, and `init`/`install-hook` write post-commit and
post-merge hooks that push every commit as it lands (warn, never block;
GITFLOW_NO_PUSH=1 for throwaway repos). T18 + T19 (installed == emitted).
- hooks/unpushed-guard.sh on SessionStart and Stop: branch ahead of its
upstream, no upstream, or no origin. Non-blocking systemMessage.
- settings.json: static deny for transfer and mirror tools, rsync --delete,
xargs rm, pipe-to-shell, chmod/chown -R, sudo/doas/pkexec, disk tools,
chattr, docker volume drops/prune/--privileged/socket/-v /:, git history
destruction, --no-verify and core.hooksPath; new hard_deny "destructive
tool against a local path, brief carries no user authority"; soft_deny
reworded + discarding uncommitted work; environment records the incident.
- CLAUDE.global.md "Destructive tools & data loss"; the four report-only
agents trace by reading, never by running, whatever the brief says.
- lib/tests/guard-bash.test.sh: executable spec of the PreToolUse guard
(214 cases). The hook itself is not shipped (BLK-022); the spec skips.
Upstream supersedes `@21st-dev/magic` with `@21st-dev/cli` (bin `21st`):
same endpoint, `21st login` in place of an API key, no MCP process loaded
into every session.
- install-plugins.sh Step 8.7: `npm i -g @21st-dev/cli` (pinned in
plugins.lock.json), staged `21st skills install`, TTY-only login offer,
pack disabled by default. update-all.sh 7.4 refreshes both.
- The documented `21st install-skill` cannot be used: the installer refuses
to follow a symlink on the target path and `~/.claude/skills` is one. The
install runs under a throwaway HOME and the result moves into
skills-external/21st-* (gitignored), symlinked on demand.
- toggle-external.sh manages `21st` as a pack (names globbed from
skills-external/21st-*, parked under plain names). `magic` is gone.
- The 5 design skills join design/web/web-full/full and MANAGED_EXTERNALS;
21st-registry and 21st-design-sync stay parked. MANAGED_MCPS is now empty
and profile.sh's dead magic branches are removed.
- Design gate: GATE-BLOCK gains `21st` (required-manual, magic's old slot)
and `21st-ui-build`; PATH repair extended to the npm global bin.
- settings.json: the 4 mcp__magic__* ask entries go; the outward-facing
21st verbs land in autoMode.soft_deny, the tier that holds under auto
mode (LRN-153).
- Docs: README, CLAUDE.global.md, design-gate.md, profile SKILL.md,
.env.example, .gitleaks.toml, link.sh. BDR-093, LRN-158.
Tests: profile-set-managed 17/17, make test green except 2 pre-existing
gitflow FAILs (gitleaks binary absent on this host), shellcheck clean.
`permissions.ask` gates nothing under `defaultMode: auto` (LRN-146,
verified live), so the ten rules that left the static tiers had no cover
left: rsync / kill -9 / killall / pkill out of deny, and python3 -c /
python -c / xargs / sed / cp / mv out of ask.
autoMode.soft_deny (7 rules) takes over what an explicit instruction
should be able to clear: writes outside the working directory,
rsync --delete, SIGKILL and kill-by-name, in-place edits spanning more
than one file, directory moves, and inline interpreters or xargs that
delete or write outside the cwd. Intent clears a soft block for the
current turn only, stated as a rule since no setting expresses it.
autoMode.hard_deny (3 rules) takes the classes no command pattern can
express: secret exfiltration, production deployment, and disarming the
guardrails. Adding a restriction stays allowed, removing one does not.
permissions.deny gains ten .env reader rules (sed awk cut tr sort uniq
diff od xxd strings). Six of those tools sat in permissions.allow, so
reading a .env through them triggered nothing.
autoMode.environment named another project, its FTP deploy target and its
customer data, inside the file link.sh:21 symlinks to
~/.claude/settings.json, where it reached every repo and contradicted
this one's Gitea remote. Rewritten machine-generic; the project facts
moved to that project's gitignored .claude/settings.local.json. All three
lists now open with "$defaults", which the original omitted, so the
built-in classifier entries are inherited rather than replaced.
doctor.sh check_automode backstops both defects. SETTINGS.md documents
the block and a tier-choice table. README no longer claims the ask tier
makes every mcp__magic__* call require a live confirmation.
- Fresh-install block: clone URL → github.com/bchanot/claude, bash
install.sh/doctor.sh → make install / make doctor (user pass)
- magic MCP example: placeholder key line instead of WRONG/RIGHT contrast
- new subsection: SEO data layer needs GOOGLE_OAUTH_CLIENT_ID/SECRET +
CRUX_API_KEY in ~/.claude/.env (GCP steps, make seo-connect, graceful
degradation) — mirrors .env.example
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.
Full removal per user request: the PreToolUse hook that blocked model
Edit/Write on quality-gate files (settings.json, gitflow.sh, .githooks,
doctor.sh, hooks, lib/tests, lint configs) plus its one-shot sentinel.
- delete hooks/config-protection.sh
- delete lib/tests/config-protection.test.sh
- deregister the hook from settings.json (rtk-rewrite PreToolUse kept)
- drop the README mention
Residual protection unchanged: gitflow pre-commit guard + Gitea branch
protection still block direct code commits to main/develop.
BDR-059 + LRN-110 + LRN-111. Confirmed A's ask-gate covers component_builder
(mcp__ scope) — no code fix possible or attempted, it's third-party package
code (dist/utils/callback-server.js:36). README MCP section now documents
the risk and why the mitigation is ask-gating, not patching.
toggle-external.sh's `claude mcp add magic --env API_KEY="$MAGIC_API_KEY"`
materialized the key as plaintext into ~/.claude.json — a copy outside the
~/.claude/.env canonical, invisible to the repo's gitignore/allowlist reach.
Claude Code supports ${VAR} expansion in mcpServers config (docs confirmed),
so the fix is a reference, not a scrub.
- lib/toggle-external.sh: --env 'API_KEY=${MAGIC_API_KEY}' (single-quoted
literal reference, not bash-expanded) so future `enable magic` runs write
the safe form too.
- README: new "Adding an MCP server that needs a secret" section documenting
the --env pitfall and the wrapper pattern.
Out-of-repo companion changes (not in this commit): ~/.bashrc gained a
scoped claude() wrapper that sources ~/.claude/.env into a subshell before
exec'ing the real binary (verified: the var never reaches the ambient
interactive shell, only claude + children) — chosen over a global export to
keep the secret's surface minimal. ~/.claude.json's mcpServers.magic.env.API_KEY
was rewritten to the same "${MAGIC_API_KEY}" reference via a surgical jq
edit (never read directly, so the value never entered this session's
context). The 2 of 5 rotating ~/.claude/backups/.claude.json.backup.* files
still holding the old plaintext were scrubbed the same way.
Residual: this session predates the bashrc wrapper, so `claude mcp list`
currently warns "Missing environment variables: MAGIC_API_KEY" — expected,
resolves on next terminal + Claude Code restart. MAGIC_API_KEY rotation
still pending (user action, after this commit).
- README + plugins.lock.json graphifyy note: pipx/PyPI install only, never
npm/npx — a different publisher (rhanka/graphify) squats the same
'graphifyy' name on npm as a version-shadowing shim with its own
conflicting 'graphify' bin (F-X1).
- agents/security-auditor.md: one-line caveat that p/* semgrep packs are
fetched from the registry at runtime — the CLI version pin does not
freeze ruleset content, so a new BLOCK can appear on unchanged code.
MCP magic (F-X3): version pin declined by user call (stays @latest in
lib/toggle-external.sh). ${VAR} env expansion confirmed unsupported at
~/.claude.json user scope (Claude Code docs — expansion is .mcp.json
project-scope only), so the BDR-026 reference-not-plaintext pattern
doesn't transfer here; existing mitigations (canonical ~/.claude/.env,
gitignore, audit env-field filtering) remain the practical ceiling.
~/.claude.json regenerated out-of-repo via toggle-external.sh disable+
enable magic to pick up the already-rotated MAGIC_API_KEY (no repo diff,
no commit for that file — traced in the job6 final report).
Complementary to frontend-design (kept: build-time aesthetic direction).
impeccable adds what the chain lacked: 45 deterministic anti-slop rules
(npx impeccable detect, exit 0/2, --json) — the design counterpart of the
semgrep gate — plus 23 design verbs under one /impeccable skill and
persistent per-project design context.
- plugins.lock.json: CLI pinned 3.2.0 (rules update = audit output change
on unchanged code, LRN-077 class); skill dist = its own release track
- install-plugins.sh Step 8d: staged npx install (tmpdir) -> moved to
skills-external/impeccable (machine-owned, gitignored, ctx7 pattern);
never writes through the ~/.claude/skills symlink into the tracked tree
- update-all.sh: pin-honored refresh, Node<24 or failure -> dist kept
- Node >= 24 required (host at 22): steps skip gracefully, activation
deferred to a deliberate Node bump
- link.sh EXTERNAL_SKILLS, profiles (design/web/web-full/full),
plugin-advisor, CLAUDE.md design routing, design-gate, README, CHANGELOG
- NOT in design GATE-BLOCK yet: promotion after first dogfood
Reorder: DOC SYNC moves from STEP 9 (post-FINISH) to STEP 8 (pre-FINISH); FINISH → 9. doc-syncer PATCHES public docs but never commits them, and finishing-a-development-branch integrates only committed history — so on the push+PR path the patched docs were uncommitted and never reached the PR (twin of the capitalize PR-strand fix, BDR-034). STEP 8 now chains lib/doc-commit.md: surgical commit of the PATCHED_FILES, never git add -A, never .claude/ (rc 4 loud anomaly), no-op if nothing patched. Twin-chantier HTML comment removed — the twin is fixed, 'deferred' is now false.
Ref-coherence (part of the reorder — a swap flips meanings, it doesn't just drop a number): STEP 7 capitalize body refs 'STEP 8 FINISH' → 'STEP 9 FINISH' (lines 159, 189). README.md pipeline illustration completed — STEP 4-7 ends in capitalize (not 'finish'), STEP 8 = sync README, STEP 9 = finish; it had been silently wrong since e8eff7e moved DOC SYNC 8→9 without updating it, so completed rather than left accidentally-correct. Historical records (BDR-034, journal, CHANGELOG) left as-is — they said 'STEP 8' when it was true; append-only.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ho5EQCFTSvYamuRtVZpp2d
Clearer scoped name for the W3C + WCAG skill. Updated: folder (git mv),
frontmatter name, H1 title, command refs, CLAUDE.md routing, 6 profiles
(functional — activate the skill by folder name), cross-refs in
harden/seo/depth-matrix/client-handover, agent dispatch refs, README +
USAGE tables.
Confidentiality: the client-deliverable leak-guard regex
(client-handover-writer.md) now matches BOTH /web-validate and legacy
/validate, so older client docs stay covered.
Left intentionally: validator-analyzer agent name (lockstep with
subagent_type + registry), .validate-cache/ + VALIDATE.md (audit-file
family {SEO,GEO,HARDEN,CSO,VALIDATE}.md), .claude/ history (append-only),
CHANGELOG old entry (added a new "renamed" entry instead). NL trigger
keywords kept so "validate" still routes here. Third-party html-validate
untouched.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W9sqAwZxBMZSynZoVrEJhd
Reflect the /close → /capitalize --ritual merge in the session-loaded routing
table (CLAUDE.md) and the README/USAGE command tables: capitalize now also
reconciles .claude/tasks/TODO.md, --ritual adds the end-of-session reflection,
and /close is documented as an alias. Does not touch the in-progress caveman
purge edits in these files (left unstaged).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X3e8LaH2vymmxyh36h3jFU
Run /doc clean over the public docs and apply the verified items.
- README: drop the unused CONTEXT7_API_KEY step from fresh-install (the
var is never read; add an optional ctx7 setup section), complete the
Makefile target table (+link, onboard, profile×3), add personal-skill
rows (audit-delta, capitalize, prune-memory, pdf-translate) with a
scope note, link CHANGELOG.md.
- USAGE: add audit-delta / capitalize / prune-memory to the decision
tree, the quick-decision table, and the command reference, plus a
scope note about plugin/marketplace skills.
- CHANGELOG: add a Keep-a-Changelog [Unreleased] section with a drafted
(review-flagged) summary of changes since 3.4.0.
GSD notation left unchanged: gsd --help confirms both 'gsd <cmd>' (CLI)
and '/gsd <cmd>' (in-TUI) are valid, so no blanket rewrite. CHANGELOG
[Unreleased] is a draft for human review, not exhaustive.
Co-Authored-By: Claude <noreply@anthropic.com>
Wires JuliusBrussee/caveman into the always-on tier alongside
security-guidance and superpowers. Caveman compresses Claude's output
tokens (~75%) by speaking like a caveman while keeping technical
substance. Three layers:
1. Plugin (caveman@caveman, marketplace JuliusBrussee/caveman)
— adds /caveman, /caveman-commit, /caveman-review, /caveman-stats,
/caveman-help, /cavecrew, /compress + 3 cavecrew agents +
SessionStart/UserPromptSubmit hooks from the plugin path.
2. Standalone hooks (statusline + stats badge) deployed by
caveman's own hooks/install.sh into ~/.claude/hooks/. Paths in
settings.json normalized to ~/.claude/hooks/... so this user's
home dir doesn't leak across machines.
3. caveman-shrink MCP proxy — NOT auto-registered. The bare proxy
fails health checks because it requires an upstream MCP server
to wrap. install-plugins.sh STEP 5.5 prints a snippet showing how
to register a wrapped entry (e.g. caveman-shrink-fs) when the user
decides which upstream to compress.
New helper enable_plugin() for explicit always-on activation —
'claude plugin install' only copies into cache, doesn't write
enabledPlugins. Idempotent via Python json check.
doctor.sh adds detect_caveman / detect_caveman_hooks / detect_caveman_shrink
checks plus a 300t passive-cost adder. update-all.sh refreshes hook
files via the upstream installer's --force mode.
.gitignore covers caveman runtime files materialized into hooks/
because ~/.claude/hooks is symlinked to this repo.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Context7 requires a free account + API key — added install step
and linked to context7.com in the components table.
Co-Authored-By: Claude <noreply@anthropic.com>
Bump version to 3.4.0. Changelog covers 9 new skills, 7 new agents,
install.sh bootstrap, hooks, and plugins.lock additions. README and
USAGE updated with new skill table, decision tree, and patterns.
TODO items marked done.
Co-Authored-By: Claude <noreply@anthropic.com>
bastien