The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill contains a 'Reverse mode' designed to document existing systems by reading project artifacts. This process involves ingesting untrusted data from the local environment which could potentially contain malicious instructions.\n
Ingestion points: The agent reads graph nodes, tool definitions, interrupt points, and existing documentation files (SKILL.md, docs/context/system-reverse-spec.md, graph-spec.md) as described in the Reverse Mode section of SKILL.md.\n
Boundary markers: The skill incorporates human-in-the-loop checkpoints, requiring the agent to present 'inferred' mappings for user confirmation or correction before finalization.\n
Capability inventory: The skill is authorized to write architectural decision records and defense packs to the docs/specs/ directory using file system tools.\n
Sanitization: Analysis is performed through discrete interview steps and a structured reporting template, providing a constrained output format for external data processing.\n- [COMMAND_EXECUTION]: The skill executes a localized Python utility, scripts/scaffold-report.py, to generate the initial structure for architecture decision records (ADR).\n
Evidence: SKILL.md instructions specify running the script to extract ADR skeletons from templates.\n
Context: The script is a utility for extracting markdown skeletons from a local template file and writing them to the project's documentation directory.\n- [DYNAMIC_EXECUTION]: The skill utilizes a local script for runtime file generation based on project templates.\n
Evidence: The scripts/scaffold-report.py script is used to dynamically assemble the structure of the agent-architecture.md file at runtime.

ns-agent-architecture