The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive shell operations using git (rev-parse, show, tag, worktree, switch, status, log) and nvidia-smi to manage and inspect the repository environment. It also executes repository-local build scripts like build_lib.py via uv run.
[REMOTE_CODE_EXECUTION]: In Phase 3a, the skill is instructed to generate standalone Python scripts in /tmp/ based on claims found in the CHANGELOG.md. These scripts are then executed using uv run. This creates a runtime code execution capability where the executed logic is derived from external text entries.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).
Ingestion points: Reads CHANGELOG.md bullets, VERSION.md, and warp/config.py as primary inputs (SKILL.md Phase 1, Phase 2, Phase 3).
Boundary markers: None explicitly mentioned or implemented for isolating or delimiting changelog content during analysis.
Capability inventory: Performs subprocess calls for git, gh, and uv run. It has file write access to CHANGELOG.md and generates temporary scripts in /tmp/ (SKILL.md Phase 3a, Phase 5, Phase 6).
Sanitization: No specific sanitization, escaping, or validation of changelog content is described before it is used to generate verification scripts or rewritten prose.

changelog-audit