The Agent Skills Directory

[SAFE]: The skill acts as a local documentation browser and command reference tool. It does not perform network operations, exfiltrate data, or execute arbitrary code.
[DATA_EXPOSURE]: The skill is designed to read project state files (e.g., .workflow/state.json) and documentation files (e.g., ~/.maestro/workflows/) using the Read, Grep, and Glob tools. This behavior is documented and necessary for its purpose of providing context-aware help and command details. Access is limited to the local filesystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text through the $ARGUMENTS variable to determine the help mode and search keywords. While this allows user input to influence the agent's behavior (e.g., which documentation to show), the risk is minimal as the skill only outputs informational text and lacks capabilities to perform destructive actions or reach external servers.
Ingestion points: User input from $ARGUMENTS is processed in phases/01-parse-intent.md to route to different operation modes.
Boundary markers: The logic utilizes distinct processing phases (Parse, Search, Present) to separate intent detection from data retrieval, though it does not use explicit character delimiters for user input.
Capability inventory: Restricted to Read, Grep, Glob, and AskUserQuestion tools.
Sanitization: Intent parsing relies on keyword matching and verification against a static list of known commands provided in the index/catalog.json file.

maestro-help