The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core function involves processing and reviewing arbitrary code changes, which serves as a significant surface for indirect prompt injection. Untrusted data from code diffs is piped directly to various LLMs.
Ingestion points: Code diffs and untracked file contents are read using git diff and git ls-files in SKILL.md, headless-claude.mjs, headless-codex.mjs, and headless-cursor.mjs.
Boundary markers: The instructions do not specify the use of clear delimiters or 'ignore' directives to prevent the model from obeying instructions embedded within the code being reviewed.
Capability inventory: The skill can modify the local repository state (git apply), manage processes, and execute external CLI tools.
Sanitization: There is no evidence of escaping or filtering of the repository content before it is passed to the models.
[DYNAMIC_EXECUTION]: The skill utilizes several Node.js wrapper scripts to dynamically orchestrate the behavior of external CLI tools based on the review state.
Execution patterns: Scripts such as headless-claude.mjs, headless-codex.mjs, and headless-cursor.mjs construct command arguments at runtime and execute them via the supervise utility.
High-privilege flags: The headless-claude.mjs script explicitly invokes the claude CLI with --permission-mode bypassPermissions to facilitate automated code simplification edits without per-change manual approval.
[COMMAND_EXECUTION]: Extensive use of system-level commands to manage environment state and repository history.
Git operations: The skill executes numerous git commands to resolve merge bases, manage local history, and fingerprint the working tree to ensure integrity.
System inspection: The run-transport.mjs script executes /bin/ps to perform process ancestry checks for environment validation.
Process orchestration: The skill interacts with the herdr CLI to manage visible execution panes and execute commands within remote shell contexts.
[EXTERNAL_DOWNLOADS]: The skill instructions specify a dependency on functionality provided by a separate, external skill.
Unverifiable dependency: SKILL.md directs the agent to execute node scripts/usage-state.mjs from a sibling skill named 'orchestrate'. This dependency is external to the reviewed codebase and its behavior cannot be verified within this scope.

review-it