The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from .claude/agents/activity/{branch}.jsonl to automatically populate 'Agents Involved' and 'Co-Authored-By' sections in commit messages.
Ingestion points: Phase 3b reads the activity ledger for the current branch.
Boundary markers: None present; the skill treats content from the JSONL as trusted names and descriptions for attribution.
Capability inventory: The skill uses git commit to create new history and Write to create handoff files.
Sanitization: The skill filters entries based on timestamp and duration but does not sanitize the string content of agent names or activity descriptions.
[COMMAND_EXECUTION]: The skill performs multiple shell operations via the Bash tool to validate code and manage the git lifecycle.
Evidence: Executes git, npm run, and poetry run commands in Phase 2 and Phase 4 to perform linting, testing, and committing. It also includes a scripts/validate-conventional.sh utility for message validation.
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes invocation_hooks to execute shell commands during the initialization phase.
Evidence: invocation_hooks includes a git rev-parse command to verify the git environment when the skill is loaded.
[DYNAMIC_EXECUTION]: The skill uses a PreToolUse hook to execute an external script at runtime.
Evidence: The YAML frontmatter contains a PreToolUse hook that runs ${CLAUDE_PLUGIN_ROOT}/hooks/bin/run-hook.mjs. This script is used to load configurations but represents a dynamic execution point.

commit