The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform a variety of system operations as part of its simulation and audit pipeline. This includes using git to autonomously create branches, stage changes, and commit fixes for identified high-priority issues (M-tier). It also executes several external CLI tools for multi-persona analysis, such as ollama, gh copilot, and agy. The skill also employs perl to create a portable timebox for command execution, which involves the use of the exec function.
[REMOTE_CODE_EXECUTION]: The skill utilizes npx @openai/codex as part of its analysis workflow. Additionally, the skill's 'Persona Discovery' process includes a mechanism to query a /plugin-recommender for missing persona agents and can execute an 'install command' to add plugins from external or marketplace sources. While this process is intended to involve user confirmation, it represents a path for downloading and executing arbitrary code.
[EXTERNAL_DOWNLOADS]: As part of its orchestration of parallel agents, the skill can identify gaps in its persona set and download additional plugins from external marketplaces or repositories based on suggestions from the /plugin-recommender tool.
[INDIRECT_PROMPT_INJECTION]: The skill audits and processes external artifacts including SKILL.md files, READMEs, and source code. These files are ingested and passed to sub-agent personas for evaluation, creating a surface for indirect prompt injection if a processed file contains malicious instructions. Given the skill's autonomous capabilities to modify and commit code changes, a successful injection could lead to unauthorized repository modifications.
Ingestion points: Reads target artifacts such as code and documentation into the agent context via the Read tool.
Boundary markers: Missing; the skill does not use specific delimiters or instructions to ignore potential commands embedded within the processed artifacts.
Capability inventory: Includes the Write tool for fixing issues and the Bash tool for committing changes to the repository.
Sanitization: There is no evidence of sanitization or escaping of the ingested artifact content before it is processed by the persona agents.

sim-conductor