The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands using external CLI tools, specifically codex exec and claude -p. Additionally, it requires the execution of arbitrary commands for project verification, such as running "tests, a build, or behavior checks."
[COMMAND_EXECUTION]: The skill contains instructions for the agent to self-modify its own source file (SKILL.md). Specifically, it mandates: "If you discover the CLI exposes a newer/stronger model, edit this skill to update these defaults."
[PROMPT_INJECTION]: The skill creates a surface for Indirect Prompt Injection because it iterates on arbitrary artifacts (code, design, documents) that may contain malicious instructions.
Ingestion points: Ingests untrusted data through the artifacts provided for review and hardening.
Boundary markers: The skill does not define clear delimiters or "ignore" instructions for the processed content, deferring the logic to external sub-skills.
Capability inventory: The agent possesses high-impact capabilities including shell command execution (codex, tests, build) and file modification (self-editing the skill file).
Sanitization: There is no evidence of sanitization or filtering of the artifact content before it is processed by reviewers or executed in test environments.
[REMOTE_CODE_EXECUTION]: By orchestrating external agents via codex exec, the skill facilitates the remote execution of tasks which may be influenced by instructions embedded within the processed artifacts.

converging-and-polishing