The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands using git and gh CLI tools to gather repository metadata, resolve version strings, and manage GitHub Gists.
[COMMAND_EXECUTION]: Included Python scripts (scripts/list_commits.py and scripts/diff_public_api.py) are run via uv run to perform deterministic data extraction and API surface analysis. These scripts utilize subprocess.run to call Git internally.
[COMMAND_EXECUTION]: In Phase 4g, the agent is instructed to verify ambiguous breaking changes by generating and executing minimal Python test scripts in a temporary directory. This involves dynamic code generation and execution based on analyzed repository content.
[DATA_EXFILTRATION]: The skill facilitates the transfer of generated audit reports to an external service (gist.github.com). This is the intended output method and leverages the user's authenticated gh session to target a well-known, trusted service.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from commit messages and CHANGELOG.md to inform its analysis and reporting.
Ingestion points: Data enters the context via git log, git show, and file reads of CHANGELOG.md (specified in Phase 2 and Phase 3).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present when the agent processes this external content.
Capability inventory: The skill utilizes subprocess.run, uv run, and file writing capabilities, along with network access via the gh tool.
Sanitization: There is no evidence of sanitization, validation, or escaping of the repository data before it is interpolated into the agent's prompts or used to guide script generation in Phase 4g.

release-audit