goal-quench
Warn
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPERSISTENCEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses Bash scripts to perform environment checks, manage internal state files, and resolve verification artifacts using git commands.
- [REMOTE_CODE_EXECUTION]: Fetches and executes the '@chrono-meta/fh-gate' utility from the npm registry using npx for post-run quality governance.
- [PERSISTENCE]: Instructs users to modify '.claude/settings.json' to install a 'Stop-hook,' which persists the skill's verification behavior across future agent sessions.
- [DYNAMIC_EXECUTION]: Generates and executes a Python script at runtime to parse session transcript files (JSONL) for token usage analysis and calibration.
- [DATA_EXFILTRATION]: The skill accesses sensitive platform files, including '.claude/settings.json' and conversation transcript logs ('.jsonl' files) located in the agent's project directories.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided task descriptions and external search results, creating a surface for embedded instructions to influence agent behavior.
- Ingestion points: Phase 1 Step 1 (SKILL.md) collects task descriptions; Phase 1.5 Step C (SKILL.md) performs external discovery using WebSearch and WebFetch tools.
- Boundary markers: Absent for both task descriptions and external search results.
- Capability inventory: The skill utilizes Bash, Write, and Agent tools across both SKILL.md and SKILL_detail.md.
- Sanitization: No explicit sanitization or filtering of external content is defined before interpolation into prompts.
Audit Metadata