The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes data from memory/logs/ and output/articles/ to define the functionality of the prototypes it builds. This creates a surface for indirect prompt injection where malicious instructions in these files could influence the generated output.
Ingestion points: Accesses memory/logs/${today}.md, output/articles/, memory/topics/*.md, and memory/MEMORY.md (SKILL.md, steps 1-2).
Boundary markers: No specific delimiters are used to separate ingested data from agent instructions.
Capability inventory: Executes shell commands (git, gh, rm), writes files to the local file system, and performs network deployment to Vercel (SKILL.md, steps 4, 6, 8).
Sanitization: Implements a 'Pre-flight check' (SKILL.md, step 6) that scans all files for sensitive patterns like VERCEL_TOKEN, ANTHROPIC_API_KEY, and various credential formats (sk-ant-, ghp_, etc.) before deployment. This effectively mitigates accidental data exposure but does not sanitize for malicious instructional content.
[COMMAND_EXECUTION]: The skill uses local shell commands for project management and deployment, including rm -rf, mkdir, git, and the GitHub gh CLI. These operations are restricted to project-specific paths (.pending-deploy/) and are consistent with the skill's deployment purpose.

deploy-prototype