lean-ctx-eval
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXPOSURE_&_EXFILTRATION]: The skill accesses sensitive session history and interaction transcripts stored in
~/.claude/projects/. This data contains previous prompts, code snippets, and AI responses which are processed to calculate token usage. - [COMMAND_EXECUTION]: The skill executes a local Python script (
lean_ctx_eval.py) found within the skill directory or the user's plugin folder to perform the analysis. It also creates a Python virtual environment and installs thetiktokenlibrary usingpipto facilitate accurate token counting. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from historical session transcripts.
- Ingestion points: Reads transcripts from the
~/.claude/projects/directory inSKILL.mdstep 3. - Boundary markers: None; the instructions do not specify delimiters or warnings for the agent to ignore instructions embedded within the analyzed transcripts.
- Capability inventory: Uses a Python script to perform file reads of transcripts, writes output to
/tmp, and utilizes thetiktokenlibrary for processing. - Sanitization: None; the skill does not describe any filtering or escaping of the transcript content before it is processed or relayed in the final report.
Audit Metadata