The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/proof_gate.py utility includes a run command that executes user-specified shell commands. This capability is used to invoke project-specific test runners (such as Playwright) to generate acceptance evidence. The execution is explicit and requires the caller to provide arguments directly, preventing automatic command generation from untrusted data.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes requirement data from the repository, which is a potential surface for indirect injection.\n
Ingestion points: Markdown files containing structured proof-contract JSON blocks (e.g., proof/STORY-42/acceptance.md) within the Git worktree.\n
Boundary markers: JSON content is isolated within specific fences (```json proof-contract).\n
Capability inventory: File system read/write access, Git repository metadata inspection, and arbitrary command execution via subprocesses.\n
Sanitization: The skill employs strict JSON schema validation in the validate_contract function and includes a guard_path security utility that detects and blocks symlinks or junctions to mitigate path traversal risks.

proof