The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from git commit messages and user-generated markdown reports via its checking scripts (check-commits.mjs, check-report.mjs).
Ingestion points: Commit logs from git log and report files located in the reports/ directory.
Boundary markers: The scripts utilize regular expressions to identify Conventional Commit headers, specific trailers (e.g., Gate-Change), and Markdown headings.
Capability inventory: The skill allows the agent to stage and commit code to git repositories and write files based on templates.
Sanitization: The tool performs structural validation using regex but does not perform content sanitization of the processed messages.
[COMMAND_EXECUTION]: The provided scripts (check-commits.mjs, check-tags.mjs, check-lib.mjs) execute local git commands using spawnSync and execFileSync to retrieve commit history, tags, and staged files. These operations are intended for local validation and do not involve unauthorized network communication.
[DYNAMIC_EXECUTION]: The shared library check-lib.mjs contains functions such as applyMutation and importPackage which facilitate dynamic file system operations (write, delete, rename, symlink) and dynamic imports. These utilities are primarily designed for the assembly and management of test fixtures used by the skill's self-test suite (selftest.mjs).

git-commits-and-reporting