diagnosing-superpowers
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external agent session transcripts which may contain untrusted content designed to subvert or manipulate the diagnosing agent. Ingestion points include transcript files identified in SKILL.md and references/session-discovery.md. Boundary markers are present in references/context-safety.md and prompts/analyst-common.md, explicitly instructing the agent to treat content as evidence rather than commands. The skill possesses capabilities for process execution and network access. Sanitization is performed through a multi-stage scrubbing and auditing process in prompts/scrub.md and prompts/scrub-audit.md.
- [COMMAND_EXECUTION]: The skill utilizes shell commands for file analysis (wc, awk, sed), archive creation (zip, tar), and managing GitHub issues via the gh CLI tool.
- [EXTERNAL_DOWNLOADS]: Interacts with the GitHub API and specific repositories (e.g., jnMetaCode/superpowers-zh) using curl and gh to search for and retrieve issue data. These resources are associated with the skill's author.
- [DATA_EXFILTRATION]: The skill is capable of bundling session transcripts and creating GitHub issues. To mitigate data exposure, it implements a strict redaction policy (references/redaction-policy.md) targeting API keys, tokens, personal identifiers, and proprietary terms before any data is exported or shared.
Audit Metadata