The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines sub-agents (complexity-triage, doc-reviewer, doc-updater, tdd-guide) that ingest untrusted data such as user requirements, Notion tickets, and code diffs. While the definitions include 'Input Defense Baseline' instructions to ignore embedded commands, the agents possess capabilities like 'Bash', 'Write', and 'Edit'.\n
Ingestion points: Task descriptions and requirements in complexity-triage.md, doc-reviewer.md, doc-updater.md, and tdd-guide.md.\n
Boundary markers: The 'Input Defense Baseline' sections instruct agents to treat inputs strictly as data.\n
Capability inventory: Tools include Bash, Write, Edit, Read, Grep, and Glob.\n
Sanitization: Primarily relies on LLM-based policy enforcement rather than input filtering.\n- [EXTERNAL_DOWNLOADS]: The skill documentation in SKILL.md describes a fallback mechanism to download agent definition files from the author's public GitHub repository (github.com/ashe-li/agent-skills) if they are missing from the local environment.\n- [COMMAND_EXECUTION]: Sub-agent definitions for doc-updater.md and tdd-guide.md authorize the use of the Bash tool to perform git operations and execute test suites, which are standard for their described purposes but represent a functional attack surface.

agents