The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/read_machine_context.py modifies sys.path at runtime using a path computed relative to the script's location (parents[3] / "scripts") to import the context_json module. Additionally, scripts/terminal_artifact_envelope.py performs Base64 decoding of data objects for artifact embedding.
[COMMAND_EXECUTION]: Several scripts, including scripts/snapshot_worktree.py, scripts/verify_candidate_binding.py, and tests/make_review_fixture.py, utilize subprocess.run and subprocess.Popen to execute git and python3 commands with arguments derived from the skill's operational state.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted source code (the 'Candidate') and related JSON manifests, creating a surface for indirect prompt injection.
Ingestion points: Source code files within the targeted Git repository and EVIDENCE-MANIFEST.json files.
Boundary markers: Instructions in SKILL.md mandate the use of specific summary tools and advise against full ingestion of raw data to maintain context efficiency.
Capability inventory: The skill possesses the capability to execute shell commands (via git), read arbitrary local files within the review root, and write findings to report files.
Sanitization: The skill uses scripts/read_machine_context.py to selectively read JSON fields via JSON Pointers, reducing exposure to untrusted content.
[PRIVILEGE_ESCALATION]: The tests/make_review_fixture.py script programmatically modifies file permissions using target.chmod(0o755) to set executable bits on generated test artifacts.

code-reviewer