The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process external data, such as LLM transcripts and brain dumps, which introduces a surface for indirect prompt injection.
Ingestion points: The skill ingests transcripts and drafts provided in the argumentHint or referenced as source material in SKILL.md and references/distillation-process.md.
Boundary markers: The instructions do not specify explicit delimiters or "ignore previous instruction" markers to prevent the agent from obeying commands embedded within the source text being distilled.
Capability inventory: The skill possesses the ability to read vault files, generate new markdown notes, and move files to the _distilled/ directory.
Sanitization: No explicit sanitization or filtering of the source content is mentioned in the distillation workflow.
[PROMPT_INJECTION]: The file references/canon-collaboration-standard.md contains persona-steering instructions, directing the agent to act as an "intellectual peer," "compress aggressively," and avoid "speech mode." While these instructions override default agent behavior to establish a specific tone, they are aligned with the skill's purpose as a thinking partner and do not attempt to bypass safety guidelines.
[COMMAND_EXECUTION]: The skill documentation and Python script suggest or execute local commands for file management and validation.
Evidence: The distillation-process.md file suggests using ls, grep, and mv for vault organization. The check_distillation.py script uses the Python pathlib and rglob modules to verify file existence and frontmatter consistency within the local vault environment.

write-notes