The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or safety bypasses were detected. The skill's behavior is consistent with its stated purpose of providing project retrospectives.
[COMMAND_EXECUTION]: The skill utilizes subagents to execute standard git commands (git log, git diff, git tag) and scan for project artifacts. These operations are used to collect factual context for analysis and do not involve elevated privileges or unsafe command construction.
[EXTERNAL_DOWNLOADS]: The local utility script scripts/ledger.py is used for data processing. Its metadata specifies no external dependencies, ensuring execution using only the standard library.
[DATA_EXFILTRATION]: Artifacts and reports are saved locally or pushed to the user's project repositories via git. No external network connections or unauthorized data transfers were identified.
[PROMPT_INJECTION]: The skill processes untrusted data from git logs and project files, constituting an indirect prompt injection surface. Ingestion points: Git commit history and artifact content gathered in Phase 1. Boundary markers: Phase 3 uses distinct Markdown templates for evidence records and interpretation documents to isolate factual data. Capability inventory: Local file writes, git operations, and execution of the included ledger.py script. Sanitization: Relies on structured output formatting and the agent's internal analysis logic to prevent accidental instruction obedience.

aar