The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local shell scripts and CLI tools to automate the development workflow:
scripts/skills-browser.sh: Used for visual smoke testing.
scripts/mux-worktree.sh: Used for managing git worktrees for feature isolation.
scripts/pr-unresolved-count.sh: Used as a gate to ensure all PR feedback is addressed.
bun, vitest, and gh (GitHub CLI): Used for linting, testing, and PR management. These operations are consistent with the skill's stated purpose.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it processes data from external sources:
Ingestion points: Reads PR feedback via gh and processes findings from review agents (self-reviewer, code-reviewer).
Boundary markers: No explicit boundary markers or 'ignore' instructions are used when interpolating external feedback into the agent's context in SKILL.md or REFERENCE.md.
Capability inventory: The agent has the ability to execute shell commands (via bun and local scripts) and write to the file system (e.g., creating files in .claude/rules/).
Sanitization: The skill includes a 'Security Gate' in REFERENCE.md that explicitly instructs the agent to check for eval(), innerHTML, and hardcoded secrets before proceeding with a PR. This serves as a manual/guided sanitization step.

go