The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from a codebase (source code, build manifests, and imports) to generate summaries and documentation, which could contain malicious instructions.
Ingestion points: Local repository files and manifests accessed via file-reading operations and system tools (SKILL.md, Sections 1 and 2).
Boundary markers: The skill does not define explicit delimiters or 'ignore' instructions for the code content being analyzed.
Capability inventory: Includes reading local files, executing repository-level shell commands, writing JSON files to the filesystem, and spawning sub-agents via the Agent tool for quality review (SKILL.md, Section 8).
Sanitization: The skill mandates strict naming conventions (Simplified Technical English) and a review step (Step 8), which acts as a filter but does not strictly prevent the processing of adversarial instructions embedded in comments or strings.
[COMMAND_EXECUTION]: The skill utilizes shell commands to interact with the repository environment.
Evidence: Instructs the agent to run git rev-parse --show-toplevel to identify the project root and ls to verify file existence (SKILL.md, Gate and Section 1).

atlas-write