skill-cefailures

skill-cefailures

Agent skills for specific libraries, plus tooling to create new skills from online documentation. The skills can be used from Claude Code and Codex.

Each library gets a skill that provides API/pattern knowledge and debugging/troubleshooting guidance to an AI coding agent.

Installation

As a plugin (recommended)

In Claude Code, run:

/plugin marketplace add brocef/skill-cefailures
/plugin install skill-cefailures

Skills are available as /skill-cefailures:<skill-name> (e.g. /skill-cefailures:ieee).

In Codex, add the marketplace and install the plugin:

codex plugin marketplace add brocef/skill-cefailures --ref main
codex plugin add skill-cefailures@skill-cefailures

Skills are available as skill-cefailures:<skill-name> in Codex sessions after starting a new thread.

Local development

If you've cloned the repo, you can load it directly:

claude --plugin-dir /path/to/skill-cefailures

Creating a new skill

pip install -r requirements.txt

# Using OpenAI (default — requires OPENAI_API_KEY)
python scripts/create_skill.py --name knex --url "https://example.com/knex-docs.md"

# Using Anthropic SDK (requires ANTHROPIC_API_KEY)
python scripts/create_skill.py --name knex --url "https://example.com/knex-docs.md" --backend sdk

# Using the local Claude CLI (no API key needed; requires Claude Code installed)
python scripts/create_skill.py --name knex --url "https://example.com/knex-docs.md" --backend cli

This fetches the documentation, uses an LLM to analyze and split it into a SKILL.md routing layer plus topical reference docs, and writes everything to skills/<name>/.

Backend choices:

• --backend openai (default) — uses the OpenAI API. Requires pip install openai and OPENAI_API_KEY (or --api-key).
• --backend sdk — uses the Anthropic API directly. Requires pip install anthropic and ANTHROPIC_API_KEY.
• --backend cli — shells out to the local claude CLI (requires Claude Code).

Optional flags: --model <id> overrides the per-backend default model; --force overwrites an existing skill directory.

Repo Structure

.agents/plugins/
  marketplace.json            # Codex marketplace listing
.claude-plugin/
  plugin.json                 # Plugin manifest (name, version, skills/commands paths)
  marketplace.json            # Marketplace listing
.codex-plugin/
  plugin.json                 # Codex plugin manifest
commands/
  brain-style/                # /skill-cefailures:brain-style:* commands
  broker/                     # /skill-cefailures:broker:* commands
  documentation-sync/         # /skill-cefailures:documentation-sync:* commands
  permissions-auditor/        # /skill-cefailures:permissions-auditor:* commands
  process-inbox.md            # /skill-cefailures:process-inbox
  process-inbox-initiative.md # /skill-cefailures:process-inbox-initiative (orchestrator-only)
skills/                       # Nine skills, one directory each
  brain-style/
  broker/
  capabilities-sdlc/
  documentation-sync/
  elkjs/
  ieee/
  knex/
  openai-api/
  typebox/
  <name>/
    SKILL.md                  # Routing layer (loaded on invocation)
    docs/
      <topic>.md              # Detailed reference (read on demand)
plugins/
  skill-cefailures -> ..      # Codex marketplace pointer to the repo-root plugin
scripts/
  create_skill.py             # Generate skill from URL
  analyze_permissions.py      # Analyze permission request logs
  apply_permissions.py        # Apply curated permission lists to settings.json
  log-permission-requests.sh  # Permission logging hook script
  broker_server.py            # Broker server: state, routing, persistence
  broker_client.py            # Async socket client for the broker
  broker_cli.py               # Broker CLI: server, REPL, one-shot subcommands
  broker_constants.py         # Shared protocol constants
  broker_format.py            # Message format helpers
  broker_identity.py          # Identity resolution (workspace → identity)
  broker_storage.py           # Persistence (inbox/outbox/cursors)
tests/                        # pytest suite
docs/
  release-notes/              # Per-version user-facing notes (vX.Y.Z.md + upcoming.md)
  changelogs/                 # Per-version developer changelogs
  plans/                      # Design and implementation documents
AGENTS.md                     # Project instructions for Codex and compatible agents
CLAUDE.md                     # Symlink to AGENTS.md for Claude Code compatibility
requirements.txt

Skills