The Agent Skills Directory

[SAFE]: The skill and its associated scripts perform legitimate local operations to resolve project configuration and manage draft evaluation documentation. No malicious patterns such as unauthorized network access, credential harvesting, or obfuscation were detected.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external data (failure logs and user input), creating a surface for potential instruction injection.\n
Ingestion points: Failure logs and user-provided failure area descriptions via $ARGUMENTS in SKILL.md.\n
Boundary markers: None identified to delineate untrusted log data from processing instructions.\n
Capability inventory: File system writes to create .adlc/drafts/evals/EVAL-*.md and update evals.md in SKILL.md.\n
Sanitization: No sanitization of ingested logs is specified. This risk is evaluated as safe because the skill's impact is limited to documentation generation and model-initiated tool execution is explicitly disabled via disable-model-invocation: true.

evals-specify