The Agent Skills Directory

[SAFE]: The skill functions as a defensive policy framework, establishing strict operational boundaries and prohibiting agents from modifying security guardrails or the charter itself.
[COMMAND_EXECUTION]: Mandates the use of automated verification tools such as pytest, tsc, and npm within sandboxed environments to validate code integrity before promotion.
[DATA_EXFILTRATION]: Implements a strict deny-list for sensitive internal paths, including .env files and password hashes, effectively preventing agents from accessing or exposing credential stores.
[PROMPT_INJECTION]: Establishes safety protocols for processing tasks from external sources like Linear. Ingestion points: Linear issue descriptions labeled pi-dev:autonomous. Boundary markers: Enforces constraints defined in AGENTS.md and this charter. Capability inventory: Limits agent actions to sandboxed command execution and file writes within the skills/ directory. Sanitization: Integrates with a PII redactor and employs pattern-based rejection for hardcoded credentials.

launch-charter