describe-pr

Warn

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill executes a local shell script at "$CLAUDE_PROJECT_DIR/.claude/scripts/aggregate-reasoning.sh". The source and content of this script are not defined within the skill, representing an unverified execution path.
  • [COMMAND_EXECUTION]: Step 6 instructs the agent to execute commands found in the "How to verify it" section of the PR template. If the template file (thoughts/shared/pr_description.md) or the PR diff contains malicious commands, the agent may execute them automatically under the guise of verification (e.g., make check test).
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted data from the repository.
  • Ingestion points: Reads the PR template file (thoughts/shared/pr_description.md) and the output of gh pr diff which may contain attacker-controlled content.
  • Boundary markers: None. The skill does not use delimiters or instructions to ignore embedded commands when processing the template or diff.
  • Capability inventory: File system access, shell execution (bash), and network writes via the GitHub CLI (gh pr edit).
  • Sanitization: No validation or sanitization is performed on the ingested content before it is used to generate the description or determine command execution.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 22, 2026, 07:45 AM
Security Audit — agent-trust-hub — describe-pr