The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to execute local commands such as git and qmd. These operations are performed using argument lists rather than shell strings, which mitigates the risk of command injection. The commands are scoped to the local project vault directory for repository management and search index updates.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and process raw source files from the Clippings/ and Twitter-Captures/ directories. This ingestion process creates a surface for indirect prompt injection, where malicious instructions embedded in the raw data could potentially influence the agent's summarization or behavior.
Ingestion points: Raw Markdown files located in the Clippings/ and Twitter-Captures/ directories.
Boundary markers: The workflow uses TODO(ingest) placeholders and specific Markdown headers (e.g., Summary, Key Claims) to structure generated content.
Capability inventory: The agent can write to the Wiki/ directory and perform git commit operations via the provided Python scripts.
Sanitization: While scripts utilize YAML quoting for metadata, the core text processing relies on the agent's interpretation of the external source content.
[DYNAMIC_EXECUTION]: Scripts like scaffold-source.py and wiki-pages.py generate new project files by populating pre-defined internal templates with metadata. This is a low-risk automated file generation process necessary for the skill's project-scaffolding functionality.

wiki-ingest