add-skill
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill identifies an attack surface for indirect prompt injection where untrusted data enters the agent context.
- Ingestion points: The agent reads the
.local/directory recursively and performs external web searches during the research phase (Step 2). - Boundary markers: No specific delimiters are required for the ingested research material.
- Capability inventory: The agent can write several files (SKILL.md, evals.json, README.md, CHANGELOG.md) and execute the
make checkcommand. - Sanitization: The workflow mandates cross-validation of facts against primary sources and moving cleaned data into a
references/directory. - [COMMAND_EXECUTION]: The workflow involves running
make checkto validate newly created skill files. This is a standard development practice for ensuring that the skill adheres to required formats and passes local validation tests.
Audit Metadata