Security scanners legitimately reference dangerous patterns (eval, os.system,
subprocess shell=True, etc.) inside their own regex pattern definitions and
human-readable risk/fix descriptions. Auditing the auditor itself produced
17 CRITICAL false positives — all from its own pattern table. ship-gate had
the same issue (2 CRITICALs on a check description and a variable name
called eval_findings).
Fix:
- Add 'noqa: SEC-AUDITOR' / 'auditor:ignore-line' line-suppression directive
to all three scan loops (code patterns, prompt-injection markdown,
pip/npm runtime install detection).
- Annotate the 179 pattern-definition lines in skill_security_auditor.py
(regex, risk, fix entries) and 4 cleanup shutil.rmtree calls.
- Annotate ship-gate's two flagged lines (SEC-13 check description and
eval_findings variable usage).
- Annotate SKILL.md and references/threat-model.md tables that document
attack patterns for human readers (HTML comment <!-- noqa: SEC-AUDITOR -->).
Verified end-to-end:
skill-security-auditor self-audit: 17 CRITICAL -> 0 (PASS)
ship-gate self-audit: 2 CRITICAL -> 0 (PASS)
slo-architect: PASS (0/0)
project-management WARN unchanged (no top-level SKILL.md, expected)
PR #527 (@rx4u) submitted a pre-production audit skill that was based on the
pre-#593 layout (skills directly under engineering/). After #593 landed, the
diff would have undone the entire restructure (4500+ rename ops). Re-applying
the actual new content at the correct post-restructure path.
What landed:
- engineering/skills/ship-gate/SKILL.md
- engineering/skills/ship-gate/references/checks.md
- engineering/skills/ship-gate/references/patterns.md
- engineering/skills/ship-gate/scripts/ship_gate_scanner.py
Verified:
- python3 ship_gate_scanner.py --help → OK
- python3 ship_gate_scanner.py --version → ship-gate 1.0.0
- 1671 tests pass (was 1666; +5 for ship-gate smoke + integrity)
- engineering/.claude-plugin/plugin.json: 48 → 49 skills, v2.4.2 → v2.4.3
- marketplace.json: engineering-advanced-skills entry updated to match
Closes#527.
Co-authored-by: Rajaraman Arumugam <rx4u@users.noreply.github.com>
https://claude.ai/code/session_01Dq12xJakFRxwaoU8Pqejdm
`.mcp.json` is the canonical filename Claude Code expects for plugin-bundled
MCP server configuration. The auditor's hidden-file rule was flagging it as
HIGH severity, blocking the `--strict` quality gate documented in CLAUDE.md.
Co-authored-by: FreyaFujo <172978998+FreyaFujo@users.noreply.github.com>
Closes-PR: #596
Karpathy-style review of commit 3806b9b (the prior PR commit) caught real
issues that I missed: agents weren't fully equipped per the optional but
recommended fields in the official sub-agents spec.
Changes:
- engineering/agenthub/agents/hub-coordinator.md: narrow Bash(node *) (too
broad per defense-in-depth) -> moved node into disallowedTools; add
maxTurns: 100 (orchestrators run long); add skills: agenthub:agenthub
(preload the plugin's own guidance into agent context)
- engineering-team/self-improving-agent/agents/memory-analyst.md:
add maxTurns: 30 to bound runaway analysis loops
- engineering-team/self-improving-agent/agents/skill-extractor.md:
add disallowedTools (rm/curl/wget) — agent has Write+Edit so defense-in-
depth applies; add maxTurns: 30
- engineering/karpathy-coder/agents/karpathy-reviewer.md: fix skills field
format from path-style "engineering/karpathy-coder" to spec-correct
namespaced name "karpathy-coder:karpathy-coder" (the path syntax is the
cs-* orchestrator template convention; the official sub-agents spec uses
skill names per code.claude.com/docs/en/sub-agents); add maxTurns: 30
All 6 plugin agents (4 here + 2 in playwright-pro from prior commit) +
the 1 user agent (tech-ingester) now have name + description + tools +
disallowedTools (where write-capable) + model + maxTurns. The skills:
field is set on agents that benefit from preloaded domain skill content.
Functional smoke tests post-fix:
- memory-analyst: PASS (2 turns, 25s, 24K tokens, found 1 real orphan)
- skill-extractor: PASS (0 tool uses, 34s, 17K tokens, generated correct
plan staying read-only with new disallowedTools in effect)
- karpathy-reviewer: PASS (verified in prior session, 28 tool uses)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Per https://code.claude.com/docs/en/sub-agents, agents require YAML
frontmatter with name + description, and the field is `tools:` not
`allowed-tools:` (deprecated). Bare `Bash` allows any command including
curl/wget/rm, which violates defense-in-depth.
Changes:
- engineering/agenthub/agents/hub-coordinator.md: add full frontmatter
(name, description, tools allowlist for git/python/node/Agent,
disallowedTools for rm -rf / curl / wget / git push --force, model)
- engineering-team/self-improving-agent/agents/memory-analyst.md:
add frontmatter, read-only tools (Read, Glob, Grep)
- engineering-team/self-improving-agent/agents/skill-extractor.md:
add frontmatter, write tools (Read, Write, Edit, Glob, Grep)
- engineering-team/playwright-pro/agents/test-architect.md:
rename allowed-tools to tools, add model: inherit
- engineering-team/playwright-pro/agents/migration-planner.md:
same rename
- engineering-team/playwright-pro/agents/test-debugger.md:
rename + narrow bare Bash to npx playwright / node / npm patterns,
add disallowedTools for rm / curl / wget / destructive git
- engineering/karpathy-coder/agents/karpathy-reviewer.md:
narrow bare Bash to git read-ops + python, add disallowedTools
All registered agents now load cleanly under the sub-agents spec rather
than falling through to permissive registration.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Same root cause as #587/#591 — Claude Code's runtime loader rejects
array-form skills paths like ["./content-production", "./ai-seo", ...]
even when each entry is a valid subdirectory containing SKILL.md.
`claude plugin validate` accepts them but the loader does not.
The proven canonical layout (used by self-improving-agent in #536):
<plugin>/
├── .claude-plugin/plugin.json skills: "./skills"
└── skills/
├── <skill-1>/SKILL.md
├── <skill-2>/SKILL.md
└── ...
Restructured 9 multi-skill domain plugins:
- business-growth (4 skills moved)
- c-level-advisor (28)
- engineering (36)
- engineering-team (32)
- finance (2)
- marketing-skill (43)
- product-team (12)
- project-management (8)
- ra-qm-team (13)
Also fixed standalone plugins that had root SKILL.md + ./skills/ subdir
(agenthub, autoresearch-agent, executive-mentor, playwright-pro). The
loader rejected them despite skills="./skills" because of the conflicting
root SKILL.md (compare self-improving-agent which works because PR #536
moved its root SKILL.md). Moved each root SKILL.md into ./skills/<name>/.
Restored standalone plugin folders to their original paths after the
multi-skill restructure swept them into parent skills/ directories
(marketplace.json source paths require original locations).
Removed 7 orphaned marketplace entries that pointed to skill folders
without their own plugin.json (content-creator, demand-gen,
fullstack-engineer, aws-architect, product-manager, scrum-master,
skill-security-auditor) — these were already non-functional.
Bumped patch versions on every changed plugin and synced
marketplace.json. Marketplace now lists 29 working plugins (down
from 36).
After merge: users run `/plugin marketplace update claude-code-skills`
followed by `/plugin update --all` to pick up the working layout.
Claude Code v2.1.126 rejects "skills": ["./"] (array form) at runtime
with the same "Path escapes plugin directory: ./" error as the bare
string form, despite the official plugins reference docs explicitly
showing ["./"] as a valid example. `claude plugin validate` accepts it,
so the bug is in the runtime loader, not the manifest validator.
The proven workaround (already used for self-improving-agent in #536)
is to follow the canonical layout:
<plugin>/
├── .claude-plugin/plugin.json skills: "./skills"
└── skills/<plugin-name>/
├── SKILL.md
├── scripts/
├── references/
└── ...
Applied to 21 single-skill plugins:
- engineering-team/a11y-audit, google-workspace-cli, snowflake-development
- engineering/behuman, code-tour, data-quality-auditor, demo-video,
docker-development, helm-chart-builder, karpathy-coder, llm-cost-optimizer,
llm-wiki, prompt-governance, statistical-analyst, terraform-patterns
- finance/business-investment-advisor
- marketing-skill/video-content-strategist
- product-team/agile-product-owner, apple-hig-expert, code-to-prd,
research-summarizer
Moved with each SKILL.md: scripts/, references/, assets/, templates/,
expected_outputs/. Kept at plugin root (per Claude Code requirements):
agents/, commands/, hooks/, settings.json, README.md.
Bumped patch version on each restructured plugin so /plugin update
picks up the new layout.
Refs #539
PR #587 fixed `"skills": "./"` in 35 plugin.json files but did not bump
versions. Claude Code uses version as the cache key — same version string
means /plugin update is a no-op and users keep loading the cached
plugin.json with the broken path. The /plugin UI shows "1 error" on
every plugin from this marketplace because the installed cache is stale.
Per the official version-management docs:
> If you set `version` in `plugin.json`, you must bump it every time
> you want users to receive changes. Pushing new commits alone is not
> enough.
Bump every plugin.json by one patch using max(plugin_version,
marketplace_version) as the base so no version moves backward. Sync the
new versions back into .claude-plugin/marketplace.json.
After this lands, users run `/plugin marketplace update claude-code-skills`
followed by `/plugin update --all` to pull the fixed manifests.
Claude Code v2.1.107+ rejects bare "skills": "./" with "Path escapes
plugin directory: ./". This blocks every plugin in the marketplace from
loading. Replace with the layout-appropriate form per the official
plugins reference:
- Single-skill plugin (SKILL.md at root):
"skills": ["./"] array form (per docs example)
- Plugin with skills/ subdir:
"skills": "./skills" standard subdir layout
- Multi-skill domain plugin (skills are subfolders at root):
"skills": ["./sub1", "./sub2"] explicit list of skill dirs
The multi-skill case omits "./" from the array so the index SKILL.md
at the plugin root does not register as a skill — that would create
a redundant `marketing-skills:marketing-skills` namespace.
Verified by running `claude plugin validate` against all 35 manifests.
Closes#539
Zero-dependency Node.js skill that captures full-page screenshots of web
pages via Chrome DevTools Protocol. Supports SPA scroll container
expansion, lazy-load triggering, DOM stability detection, and tiled
capture for very tall pages. Requires only Node.js 22+ and Chrome with
remote debugging enabled.
Updated the skill description to emphasize proactive usage and clarified triggers for cost optimization. Enhanced the context and steps for cost auditing, optimization, and architecture design.
Implements Karpathy's 4 coding principles (Think Before Coding, Simplicity
First, Surgical Changes, Goal-Driven Execution) as an active enforcement
plugin, not just passive guidelines. Derived from Karpathy's X post on LLM
coding pitfalls but goes far beyond the source material with automated
detection tools, a review agent, and CI integration patterns.
Differentiator vs forrestchang/andrej-karpathy-skills (prompt-only, single
SKILL.md): this version ships real tooling that DETECTS violations instead
of just documenting principles.
Plugin contents (engineering/karpathy-coder/):
- SKILL.md with `context: fork` for skill chaining
- 4 Python tools (stdlib only):
- complexity_checker.py — cyclomatic complexity, class density, nesting
depth, function length, premature abstractions (Principle #2)
- diff_surgeon.py — diff noise ratio: comment-only changes, whitespace,
style drift, drive-by refactors, quote-style swaps (Principle #3)
- assumption_linter.py — detects "just", "obviously", "should work",
vague actions, unscoped users, missing format specs (Principle #1)
- goal_verifier.py — scores plan steps 0-3 for verification quality,
flags vague criteria, checks for final verification (Principle #4)
- 1 sub-agent: karpathy-reviewer (runs all 4 principles against a diff)
- 1 slash command: /karpathy-check (dispatches the reviewer)
- 1 pre-commit hook: karpathy-gate.sh (non-blocking, warns on violations)
- 3 reference docs: karpathy-principles.md (full context + when to relax),
anti-patterns.md (10+ before/after examples), enforcement-patterns.md
(Husky, pre-commit framework, GitHub Actions CI integration)
- .claude-plugin/plugin.json manifest (v2.3.0)
- Cross-tool compatible: works with any AGENTS.md-based CLI
All 4 scripts verified: --help passes, smoke tests run correctly.
complexity_checker catches its own nesting depth. assumption_linter
correctly flags "just", "obviously", "should work". goal_verifier
correctly scores plans with/without verification steps.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Implements Karpathy's LLM Wiki pattern as a production-grade plugin. The LLM
incrementally ingests sources into a persistent, interlinked Obsidian vault —
updating entity/concept/source pages, flagging contradictions, maintaining an
index and append-only log. Knowledge compounds instead of being re-derived by
RAG on every query.
Plugin contents (engineering/llm-wiki/):
- SKILL.md with `context: fork` frontmatter for skill chaining
- 3 sub-agents: wiki-ingestor, wiki-librarian, wiki-linter
- 5 slash commands: /wiki-init, /wiki-ingest, /wiki-query, /wiki-lint, /wiki-log
- 8 Python tools (stdlib only): init_vault, ingest_source, update_index,
append_log, wiki_search (BM25), lint_wiki, graph_analyzer, export_marp
- 8 reference docs: schema, page-formats, ingest/query/lint workflows,
obsidian-setup, cross-tool-setup, memex-principles
- Vault templates: CLAUDE.md, AGENTS.md, .cursorrules, index.md, log.md,
5 page templates (entity, concept, source, comparison, synthesis)
- Worked example vault on "LLM interpretability"
- .claude-plugin/plugin.json manifest
Cross-tool compatibility: the scripts are pure Python stdlib. Only the schema
loader changes per tool (CLAUDE.md for Claude Code, AGENTS.md for Codex CLI /
Cursor / Antigravity / OpenCode / Gemini CLI, .cursorrules for legacy Cursor).
init_vault.py --tool all installs all three.
Repo-level registration:
- Commands mirrored to top-level commands/ for repo-wide discovery
- Agents mirrored to agents/engineering/ as cs-wiki-{ingestor,librarian,linter}
- .claude-plugin/marketplace.json: new llm-wiki entry + version bump to v2.3.0
- CLAUDE.md updated: 234 skills, 313 Python tools, 432 refs, 28 agents, 27 commands
Also saved (deferred): craighewitt-mattpocock reimplementation plan at
documentation/implementation/ — 4-pod proposal for building better versions
of selected skills from thecraighewitt-skills and mattpocock-skills
collections. Not executed; awaiting user confirmation on scope.
End-to-end smoke test passed: init_vault → ingest → update_index → append_log
→ wiki_search → lint → graph_analyzer → export_marp all run against a fresh
vault with real pages, wikilinks, and frontmatter.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Self-contained skill for tracking technical changes with structured JSON
records, an enforced state machine, and a session handoff format that lets
a new AI session resume work cleanly when a previous one expires.
Includes:
- 5 stdlib-only Python scripts (init, create, update, status, validator)
all supporting --help and --json
- 3 reference docs (lifecycle state machine, JSON schema, handoff format)
- /tc dispatcher in commands/tc.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Move from data-analysis/ to engineering/
- Fix 5 cross-references to use correct domain paths
- Fix Python 3.9 compat in sample_size_calculator.py
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a skill for creating CodeTour .tour files — persona-targeted,
step-by-step walkthroughs that link to real files and line numbers.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a skill for creating polished demo videos from screenshots and scene
descriptions. Orchestrates playwright, ffmpeg, and edge-tts MCPs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Based on issue #464 submission by voidborne-d. Enhanced with English-only
content (removed all Chinese), anti-patterns section, cross-references,
plugin.json, convention-compliant frontmatter, and English eval scenarios.
behuman (193 lines + reference + 8 eval scenarios):
- Self-Mirror loop: instinctive response → reflection → conscious revision
- Show mode (2.5-3x tokens) and quiet mode (1.5-2x tokens)
- 3 English examples: emotional support, life advice, personal writing
- Based on Lacan's Mirror Stage + Kahneman's Dual Process Theory
- Zero dependencies — pure prompt technique
Co-Authored-By: voidborne-d <voidborne-d@users.noreply.github.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Each skill is now individually installable:
- llm-cost-optimizer
- prompt-governance
- business-investment-advisor
- video-content-strategist
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds self-eval skill for honest AI work quality evaluation. Uses two-axis scoring (ambition x execution), mandatory devil's advocate reasoning, and cross-session anti-inflation detection via .self-eval-scores.jsonl persistence.
- Add --include-security flag to quality_scorer.py
- Default: 4 dimensions × 25% (backward compatible)
- With --include-security: 5 dimensions × 20%
- Update tier recommendation logic for optional Security
- Update documentation to reflect opt-in behavior
This addresses the breaking change concern from PR review:
the weight change from 25% to 20% would affect all existing
audit baselines. The new opt-in approach preserves backward
compatibility.
Per Anthropic docs: "Keep SKILL.md under 500 lines. Move detailed
reference material to separate files."
- browser-automation: 564 → 266 lines (moved examples to references/)
- spec-driven-workflow: 586 → 333 lines (moved full spec example to references/)
- security-pen-testing: 850 → 306 lines (condensed OWASP/attack details, moved to references/)
No content deleted — all moved to existing reference files with pointers.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Validator expects scripts in scripts/ subdirectory, not at skill root.
Moved 6 scripts to match repo convention.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
