The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external files (prd.md, idea.md, validate.md) to extract project requirements and generate documentation. This constitutes a vulnerability surface where maliciously crafted instructions embedded in these files could influence the agent's behavior.
Ingestion points: The prd-reader agent reads and parses prd.md, idea.md, and validate.md at the start of the workflow.
Boundary markers: The skill does not explicitly define strict delimiters or instruction-bypass protection when passing extracted content between subagents (prd-reader to tech-researcher to tad-writer).
Capability inventory: The skill has permissions to write files (tad.md, README.md), perform git operations (pull, rebase, push), and execute a specific local Python script if present in the user's repository.
Sanitization: No explicit sanitization, filtering, or escaping of the input document content is mentioned in the instructions.
[COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands for repository management and validation.
Evidence: SKILL.md includes shell blocks for git synchronization: git fetch, git pull --rebase, git stash, and git push.
Evidence: SKILL.md Phase 6 mentions running python3 scripts/update_readme_ideas_index.py if the script exists in the user's repository.
Evidence: SKILL.md Edge Cases mentions validating Mermaid syntax using mmdc -i tad.md -o /tmp/check.svg.
[EXTERNAL_DOWNLOADS]: The documentation (docs/README.md) provides instructions for humans to install the skill using external package managers.
Evidence: Mentions npx skills add and asm install targeting GitHub repositories for installation.

tad-generator