The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data which presents a surface for indirect prompt injection.
Ingestion points: Ingests code-under-review, specification, and diff files as primary inputs defined in the Immutable request section of SKILL.md.
Boundary markers: Implements strict boundary controls including a read_only=true flag, an isolated context window, and the requirement for a fresh reviewer context for each lane execution.
Capability inventory: Performs localized command execution via go test for validation and coordinates code execution through the codex-exec and agy-native dependencies.
Sanitization: Enforces SHA-256 digest validation for all input files (head_sha, contract, and diff) immediately before dispatch and uses nonces to ensure request-response binding.
[COMMAND_EXECUTION]: The skill includes a specific Go testing command for runtime validation.
Evidence: SKILL.md contains the command go -C cli test ./internal/ports ./internal/adapters/reviewlane_worker -run '...' used to verify the skill's logical contract and isolation properties.
Context: This is a localized developer tool invocation intended for verifying internal ports and does not appear to accept unvalidated user input into the shell environment.

pawl-review