The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes specific templates for delegating research tasks to subagents in references/delegation-context-template.md. These templates use imperative instructions to strictly control agent behavior (e.g., "You MUST produce a compliant Artifact Pyramid", "Respond with ONLY the absolute path... Do NOT respond with natural language"). While designed to ensure machine-readable outputs for pipeline automation, these reflect high-control steering directives.
[COMMAND_EXECUTION]: The skill provides local utility scripts intended for execution by the agent: scripts/pyramid-status.sh (a Bash script for structural auditing) and scripts/extract-atoms.py (a Python script for claim extraction). These scripts interact with the local filesystem to scan directories and process text files but do not involve network connectivity or privilege escalation.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and synthesize external research data, creating a vulnerability surface where malicious instructions embedded in source text files could influence the agent during the synthesis process.
Ingestion points: The scripts/extract-atoms.py utility reads and parses content from source text files provided as user input.
Boundary markers: The methodology employs a rigid directory structure (01-summary, 02-analysis, 03-dossiers) and explicit SOURCES navigation sections to logically partition data tiers.
Capability inventory: The skill performs filesystem read/scan operations and text parsing through its included scripts.
Sanitization: Content processing in extract-atoms.py relies on heuristic regex patterns for identifying claims rather than robust safety filtering of the input text.

artifact-pyramids