self-healing
Warn
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill is designed to generate and execute shell scripts and code snippets at runtime to resolve environment issues or missing project capabilities. Evidence is found in the core heal loop and artifact generation examples.
- [COMMAND_EXECUTION]: Extensive use of shell commands for diagnosis and verification. The detect-failure.sh hook script is designed to trigger these actions on command failure.
- [DATA_EXFILTRATION]: The skill instructions specifically mention environment variables and .env files as diagnostic targets. While used for local recovery, reading these files exposes sensitive credentials to the agent context.
- [PERSISTENCE]: The skill stores diagnostic artifacts and verified patches in the .learnings/ directory and promotes fixes to project-level instruction files like CLAUDE.md to influence future sessions.
- [INDIRECT_PROMPT_INJECTION]: The skill reads and acts upon historical data stored in HEALS.md without sanitization or boundary markers. 1. Ingestion points: Reads .learnings/HEALS.md. 2. Boundary markers: None. 3. Capability inventory: Shell access, file write, and configuration modification. 4. Sanitization: None.
Audit Metadata