affinage

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitHub PR comments, review bodies, and CI logs, which could serve as a vector for indirect prompt injection.
  • Ingestion points: Reads external data from GitHub API endpoints 'pulls//comments' and 'pulls//reviews', and extracts CI check summaries as defined in SKILL.md and flow-details.md.
  • Boundary markers: The skill does not explicitly define delimiters for external content in the documentation, but it uses a structured grading and triage process.
  • Capability inventory: Accesses GitHub API to post replies, performs git merge operations, and dispatches sub-agents for automated fixes via the /cure skill.
  • Sanitization: Implements a dedicated 'reviewer' sub-agent with read-only, fresh context to analyze claims and return a compact digest to the parent context, effectively isolating the triage of untrusted data.
  • [COMMAND_EXECUTION]: The skill executes multiple command-line tools to manage repository state and resolve conflicts.
  • Evidence: Invokes 'git', 'gh' (GitHub CLI), 'mergiraf', 'rerere', and 'kdiff3' for version control and merge resolution tasks.
  • Evidence: Runs a local bundled Python utility 'affinage.pyz' for operational tasks including status checks, routing, and posting replies with attribution.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 11:39 AM
Security Audit — agent-trust-hub — affinage