The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing the canvaslms CLI tool to perform all write operations to the Canvas LMS. Instructions include commands for editing pages, assignments, and modules (e.g., canvaslms pages edit, canvaslms assignments edit).
[COMMAND_EXECUTION]: The skill includes a local Python script scripts/canvas_snapshot.py used to generate JSON backups of Canvas content. The instructions demonstrate how to identify the correct Python interpreter using the canvaslms shebang to ensure dependencies like canvasapi and canvaslms are available.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the Canvas LMS API (via CLI output and JSON snapshots). This represents an attack surface where malicious content stored in a Canvas page could attempt to influence the agent's behavior. However, the skill provides rigid rules for data handling (e.g., identity matching via regex, authoritative front matter semantics) and emphasizes technical verification (JSON diffs) which mitigates this risk. Mandatory Evidence Chain:
Ingestion points: Data enters the agent via canvaslms view commands and the scripts/canvas_snapshot.py output in SKILL.md and references/module-shape-example.md.
Boundary markers: None explicitly defined in instructions, though structured JSON format provides implicit boundaries.
Capability inventory: The skill can write to the local file system (via shell redirects) and to the Canvas API (via the canvaslms CLI).
Sanitization: Not explicitly mentioned, but operations are mostly structural (moving/renaming items) rather than processing the natural language content of the pages.
[SAFE]: Credentials for the Canvas API are managed externally by the canvaslms CLI tool. The provided script scripts/canvas_snapshot.py calls load_credentials() from the canvaslms package, which is the standard and safe method for retrieving stored tokens without hardcoding them.

canvas-course-repo