change-init
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill mines untrusted data from git commit messages and external issue trackers (GitHub, GitLab, JIRA) to reconstruct development rationale.
- Ingestion points: Git commit subjects and bodies retrieved via git log; issue titles and descriptions fetched from GitHub or GitLab.
- Boundary markers: The skill mandates the use of a structured Markdown template for all generated records and requires explicit inline citations (SHAs or URLs) for every inferred claim.
- Capability inventory: Possesses capabilities to write files to the repository (.adlc/drafts/chdr/), execute local git commands, and perform network reads via issue tracker CLI tools.
- Sanitization: The instructions require the agent to summarize external content rather than copying it verbatim and to explicitly mark the confidence level of its inferences to prevent the ingestion of malicious or deceptive instructions embedded in commit history.
- [COMMAND_EXECUTION]: Executes several local development tools including git, gh (GitHub CLI), and glab (GitLab CLI) to perform history mining and context retrieval.
- [EXTERNAL_DOWNLOADS]: Fetches issue context and metadata from established services such as GitHub and GitLab to enrich the generated documentation.
Audit Metadata