The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The security audit subcommand in references/rules/audit.md fetches repository diffs using git diff and passes the resulting ${DIFF_CONTENT} to a subagent (codex:codex-rescue) for independent review. The untrusted diff content is interpolated into the subagent's prompt without explicit boundary markers or delimiters, creating a potential surface for indirect injection where malicious code in the diff could attempt to influence the subagent's analysis.
Ingestion points: references/rules/audit.md (via git diff output).
Boundary markers: Absent for the diff content.
Capability inventory: File writes (RELEASES.json, .env.version), Git command execution, and GitHub CLI operations.
Sanitization: None performed on diff content before interpolation.
[COMMAND_EXECUTION]: The skill relies on extensive use of shell commands (git, gh, yarn, osv-scanner) to manage the release lifecycle. This involves direct interaction with the repository's version control and local configuration files.
[EXTERNAL_DOWNLOADS]: The audit workflow performs dependency installations via yarn install and security scans via osv-scanner, which involve network requests to package registries and vulnerability databases. These operations target the project's own yarn.lock and are part of standard audit procedures.

1k-bundle-release