The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically generate and execute Bash or Python scripts to inspect the keys and structure of YAML/JSON documents and log files. This is used to create bounded projections of large datasets into the agent context.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from various sources, including transcripts, logs, and external URLs. This creates a risk where instructions embedded in those sources could maliciously influence the agent's 'capture' and 'routing' logic.
Ingestion points: Data enters the context from source_refs (URLs and paths), tool outputs, transcripts, and user-provided 'brain dumps' in SKILL.md.
Boundary markers: The skill implements a fidelity scoring system but lacks explicit delimiters or warnings to ignore instructions embedded within ingested source material.
Capability inventory: The workflow includes the ability to write artifacts to the file system, execute Bash/Python code, and route tasks to other active skills (e.g., /exec, /propose).
Sanitization: While structured data is analyzed using parsers, there is no evidence of sanitization for natural language instructions found in ingested logs or transcripts.
[COMMAND_EXECUTION]: The workflow utilizes shell environments and scripting to reconcile conversation state with local file-based artifacts and manage workstream metadata.

capture