The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data from project codebase files and manifests, creating a potential surface for indirect prompt injection attacks.
Ingestion points: The agent is instructed to read project manifests (such as package.json, pyproject.toml, and go.mod) and the complete source code during the mapping and review stages (Steps 0, 2, and 3).
Boundary markers: The instructions do not specify the use of delimiters or provide explicit 'ignore embedded instructions' warnings for the ingested data.
Capability inventory: The agent is granted the ability to read files, invoke specialized tools (/review, /security, /gate), and perform write operations including landing patches and opening pull requests (Step 7).
Sanitization: The skill does not include specific requirements for sanitizing, escaping, or filtering the content ingested from the project files.
[SAFE]: The skill's instructions are focused on best practices for security engineering, observability, and code maintenance. No evidence of data exfiltration, hardcoded credentials, malicious persistence, or obfuscation was found. The logic for refuting findings before acceptance (Step 2) and verifying fixes through a gate (Step 7) provides a layer of protection against accidental or malicious modifications.

harden