The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from git logs and diffs to generate PR/MR descriptions and release notes, creating a potential surface for indirect prompt injection from repository content. * Ingestion points: Untrusted content from git history (git log), workspace changes (git diff), and existing PR metadata (gh/glab view) are ingested into the agent context as seen in SKILL.md and references/mr-pr.md. * Boundary markers: The skill mandates using specific templates (assets/*.md) and a neutral voice to structure output. * Capability inventory: The skill can execute git commands (commit, push, tag) and host CLI commands (PR/release creation/update) as detailed in references/host-cli.md. * Sanitization: The skill implements a mandatory "read-back" verification step where it re-fetches written content and compares it using diff against the intended local file to detect silent failures or injection.
[COMMAND_EXECUTION]: The skill utilizes standard developer tools (git, gh, glab) for its core functionality and provides detailed guidance on avoiding CLI traps like non-interactive shell hangs and shell-quoting hazards in references/host-cli.md.
[SAFE]: The skill includes explicit safety instructions to skip and warn about sensitive files like .env, credentials, and tokens during the staging process in references/commit.md, and forbids pushing to protected branches in the "Iron Law" section of SKILL.md.

commit-and-push