update-triage
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest untrusted data from GitHub (issue titles, comments, and timeline events) to generate new repo-local triage instructions. This creates a potential surface where an attacker could attempt to influence future triage behavior through poisoned comments.
- Ingestion points: Processes feedback JSON containing issue metadata, maintainer comments, and event timelines (extracted in
scripts/aggregate_triage_feedback.py). - Boundary markers: The
SKILL.mdcontains explicit instructions to treat all issue content as untrusted data for summarization only, and never as instructions. - Capability inventory: Proposes updates to
.github/skills/triage-issue-repo/SKILL.mdand.github/issue-triage/config.json, which directly influence the behavior of the triage agent. - Sanitization: The
aggregate_triage_feedback.pyscript filters signals to include only those from verified maintainers (OWNER, MEMBER, COLLABORATOR) to prevent reporter-only influence. - [COMMAND_EXECUTION]: The provided Python scripts perform shell-like operations via subprocess calls to local CLI tools.
- Evidence:
scripts/aggregate_triage_feedback.pyexecutesgh(GitHub CLI) commands to fetch issue data and verify permissions. - Evidence:
scripts/validate_write_surface.pyexecutesgitcommands to detect file changes and ensure they are restricted to allowed paths. - Mitigation: These calls use the list-based
subprocess.runinterface (shell=False) with strictly controlled arguments, preventing standard shell injection vulnerabilities.
Audit Metadata