The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The workflow defined in RELEASE-NOTES.md creates a surface for indirect prompt injection by ingesting and processing untrusted data from external sources.\n
Ingestion points: Steps 3 and 5 in RELEASE-NOTES.md use the GitHub CLI (gh api, gh pr view) to fetch commit messages, PR titles, and PR descriptions from external repositories.\n
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions to prevent the agent from following commands that might be embedded in the ingested commit or PR text.\n
Capability inventory: The skill is designed to read data from the GitHub API and write results to the local filesystem (e.g., releases-docs/releases/).\n
Sanitization: No sanitization, escaping, or validation logic is present to handle the untrusted external content before it is processed by the AI.\n- [COMMAND_EXECUTION]: The skill includes reference materials that instruct the agent or user to execute various shell commands for repository maintenance and environment setup.\n
Evidence: references/automation-scripts-toolkit.md contains commands for deleting directories (rm -rf node_modules .next), pruning branches (git fetch --prune), and managing stacked PRs (gt restack).\n- [EXTERNAL_DOWNLOADS]: The automation toolkit utilizes external package managers and runtimes to execute DevOps tooling.\n
Evidence: references/automation-scripts-toolkit.md references the use of npx to download and execute semantic-release and gitleaks. It also includes instructions involving the bun runtime and the gt (Graphite) CLI tool.

git-automation