The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill utilizes standard, read-only git commands (git log, git tag) to retrieve metadata from the local repository. These operations are restricted to gathering versioning and commit data.
[DATA_EXPOSURE & EXFILTRATION] (SAFE): The skill accesses local files like CHANGELOG.md for reading and writing documentation. No network egress patterns (such as curl or fetch) or access to sensitive directories (e.g., .ssh or .aws) were detected.
[PROMPT_INJECTION] (SAFE): Although the skill processes untrusted data from git commit messages (Category 8: Indirect Prompt Injection surface), it does so within a highly structured workflow that mandates summarization into a specific Markdown format. This serves as a natural constraint against the execution of malicious instructions embedded in commit messages.
Ingestion points: CHANGELOG.md and git log output.
Boundary markers: Not explicitly defined, but the 'Keep a Changelog' schema provides structure.
Capability inventory: File writing to documentation paths.
Sanitization: The instructions explicitly tell the agent to 'review' and 'categorize', effectively filtering the raw input.

docs-changelog