The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard Unix utilities including cat, jq, mkdir, cp, find, bc, grep, and test to navigate the file system, validate JSON schemas, and construct new skill directories.
[DATA_EXPOSURE]: Accesses local workspace files such as skill-plan.json, knowledge/*.json, and procedures.json to extract the metadata and content necessary for skill synthesis.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests data from multiple source files and interpolates them into prompt templates.
Ingestion points: Data is sourced from skill-plan.json, knowledge/*.json, and chapters/*/procedures.json files within the project bundle.
Boundary markers: No explicit delimiters or instructions are used to separate ingested data from the generated instructions in the resulting skill files.
Capability inventory: The skill can create new file structures and instructions through shell-based file manipulation commands.
Sanitization: The skill validates that the input plan is approved and follows a specific JSON schema, but does not perform safety sanitization on the semantic content of the procedures or concepts being processed.

skill-synthesizer