The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes human-provided feedback from a local queue to modify project source files, which creates a surface for indirect prompt injection.
Ingestion points: Reads feedback from .annotations/inbox.jsonl (specifically the comment and anchorText fields).
Boundary markers: Absent. The skill instructions do not specify delimiters or safety warnings for interpreting the external feedback.
Capability inventory: Executes shell commands via the promptbook CLI (annotations, resolve, lint, eval) and performs file-write operations to rules/*.yaml, fragments/*.md, and fixtures/*.json.
Sanitization: Absent. No explicit validation or sanitization of the comment text is performed before the agent is instructed to turn the feedback into code or configuration changes.
[COMMAND_EXECUTION]: Executes several subcommands of the promptbook CLI tool to list annotations, resolve prompts, and perform linting or evaluations. These commands are part of the core functionality and are executed locally on the user's project files.

promptbook-annotations