The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands (ls, head) to search for and identify transcript files within the workspace's agent-transcripts/ directory. While these are used for locating context, shell execution always presents a surface for potential command injection if path variables are not strictly controlled by the environment.- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes agent-transcripts/*.jsonl files which contain external conversation data. This creates a surface for indirect prompt injection where malicious content in a transcript could attempt to manipulate the reviewer or synthesizer subagents.
- Ingestion points: Transcript files located in agent-transcripts/ (referenced in SKILL.md).
- Boundary markers: None present; the transcript content or digests are substituted directly into subagent prompt templates.
- Capability inventory: Includes shell command execution, subagent spawning with tool access (readonly: false), and skill modification via the create-skill tool.
- Sanitization: No sanitization or filtering of the transcript content is specified before processing.- [DYNAMIC_EXECUTION]: The skill dynamically loads instructions from external template files (references/judgment-reviewer.md, references/tooling-reviewer.md, references/divergent-reviewer.md, and references/synthesizer.md) and routes them to subagents. The behavior and safety of the subagents are dependent on the content of these external references.

reflect