The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions (found in references/process.md and references/contract.md) direct the agent to use shell commands for file management. Specifically, it uses mkdir to ensure directories exist, git mv or mv to archive source materials, and ls or git status for verification. These operations are scoped to the intended vault maintenance workflow.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external source material (e.g., chat transcripts, raw notes), creating a surface for indirect prompt injection if the source contains adversarial instructions.
Ingestion points: Source material is ingested from local files (e.g., notes/, _assets/) during the intake phase described in references/process.md Step 0.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the intake phase.
Capability inventory: The agent has the capability to read files, move files within the local filesystem, and execute the local scripts/check_essay.py validation script.
Sanitization: The skill does not implement specific sanitization of the source material, though the rigorous rewrite process and specific style gates (enforced by the Python checker) provide significant structural constraints on the final output.

write-vault-essay