The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its asynchronous review capture mechanism. Ingestion points: The skill reads <stem>-feedback.json files as described in SKILL.md (Review capture) and references/html-artifacts.md. Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from following instructions embedded within the feedback JSON. Capability inventory: The agent has permissions to write to the filesystem (planning artifacts), execute Git commands, and invoke local scripts for diagram rendering. Sanitization: The skill processes feedback items "one at a time" without explicit evidence of sanitizing external input.- [COMMAND_EXECUTION]: The skill executes local commands for version control and artifact generation. Git operations: SKILL.md (Git hygiene) instructs the agent to commit changes with specific rationales in the messages. Diagram rendering: references/html-artifacts.md specifies the use of canvas/scripts/render_diagram.mjs to render Mermaid diagrams into the HTML companion when the canvas tool is available.- [SAFE]: The skill generates visual artifacts using a self-contained HTML template (assets/overview-template.html). The embedded Javascript is minimal vanilla JS used for local storage of feedback and triggering file downloads, which does not represent a malicious execution pattern.

better-planning-prd