The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and interpret data from external configuration files (PROJECT.json and DISCOVERY.json) to determine its routing and editorial behavior, creating a surface for indirect prompt injection.
Ingestion points: The orchestrator reads state, named artifacts, and "editorial preferences" from SKILL.md, PROJECT.json, and DISCOVERY.json.
Boundary markers: The skill instructions do not define clear delimiters or "ignore instructions" markers when reading and processing these files.
Capability inventory: The orchestrator can invoke a suite of other skills (discover-presentation, structure-agenda, generate-slides, proofread-presentation) and perform Git operations (git status, git add, git commit) to modify the local environment.
Sanitization: There is no evidence of validation or sanitization of the editorial preferences before they are applied during the "unslop" pass.
[COMMAND_EXECUTION]: The skill incorporates Git-based version control, which involves the execution of shell commands to manage the project state.
Evidence: GIT_CHECKPOINTS.md provides instructions for the agent to snapshot status, stage changed files, and create commits based on the completion of different project phases.

build-presentation