k-skill-cleaner
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill relies on the
@nomadamas/k-skillpackage executed vianpxto provide instructions, file listings, updates, and to run the local helper script. These commands download and execute code from the author's official registry at runtime. - [DATA_EXFILTRATION]: The skill accesses sensitive local directories containing AI agent transcripts and logs to perform its analysis. The
scripts/k_skill_cleaner.pyscript scans directories such as~/.claude/,~/.codex/, and~/.local/share/opencode/to count skill trigger events. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external agent transcript files, which could theoretically contain malicious instructions designed to influence the agent's summary or the tool's output.
- Ingestion points: Multiple agent-specific log paths including
~/.claude/projects/**/*.jsonl,~/.claude/transcripts/**/*.jsonl, and others in the user's home directory. - Boundary markers: Absent; the script parses these files directly as JSON or text.
- Capability inventory: File system read access and command execution via the
npxutility. - Sanitization: The script uses
json.loadsand regex patterns to extract specific fields and counts, providing an aggregated summary rather than raw content, which limits injection surface.
Audit Metadata