The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads 'tunan:solution' comments from GitHub issues to verify their accuracy against the codebase. These comments represent an untrusted data source that could potentially contain malicious instructions.
Ingestion points: Documented in SKILL.md, the agent fetches comment bodies using gh api repos/{owner}/{repo}/issues/<N>/comments.
Boundary markers: The skill identifies relevant comments using the  marker and fenced YAML blocks, but it does not employ specific instructions to the agent to ignore any embedded commands within the markdown body.
Capability inventory: The skill is authorized to perform gh api PATCH and DELETE operations, modify repository labels, and commit changes to local files like CONCEPTS.md and instruction files.
Sanitization: The validate-frontmatter.py script provides syntax validation for metadata, but there is no mechanism to sanitize the natural language portions of the comments for administrative or behavioral instructions.
[EXTERNAL_DOWNLOADS]: The skill refers users to cli.github.com to install the official GitHub CLI. This is a trusted source for development tooling.
[DYNAMIC_EXECUTION]: The skill uses local Python scripts (validate-doc-claims.py and validate-frontmatter.py) for automated validation of documentation and metadata. These scripts run within the skill's environment to ensure data integrity.
[COMMAND_EXECUTION]: The skill performs shell operations using git and gh to manage issues and the local repository. These operations are restricted to specific, non-obfuscated commands necessary for the skill's core maintenance functions.

compound-refresh