The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the 'tiangong' CLI and other internal scripts using Node.js spawnSync. These commands are executed with argument arrays rather than raw shell strings, which effectively mitigates command injection risks. This functionality is intended for delegating complex auditing tasks to the underlying toolchain.
[EXTERNAL_DOWNLOADS]: The run-remote-process-review.mjs script fetches remote lifecycle data through the vendor's CLI (tiangong process list) to create local snapshots for analysis. These downloads are part of the core auditing workflow and target the vendor's own data infrastructure.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it ingests and processes external lifecycle inventory data which could contain malicious instructions.
Ingestion points: Remote process rows fetched via CLI and local build artifacts processed in scripts/run-remote-process-review.mjs and scripts/run-review.mjs.
Boundary markers: The instructions do not define explicit delimiters to separate the audited data from the agent's instructions, though they provide a structured rulebook for the agent to follow.
Capability inventory: The skill can execute local scripts and write audit logs and findings to the file system via spawnSync and writeFileSync.
Sanitization: The process-review-rules.md requires strict schema validation using ProcessSchema.safeParse and includes rules to detect and remove 'tool-authored language' patterns, which helps mitigate the influence of data-embedded prompts.

lifecycleinventory-review