The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-supplied project descriptions (briefs) and briefing answers to generate specifications and code. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the user's project requirements.
Ingestion points: The user's initial project description (brief) and subsequent answers during Phase 2 (Briefing).
Boundary markers: The skill uses a manifest.md to track requirements and employs  and  tags to delimit its edits in project files.
Capability inventory: The skill is empowered to write files, execute git commands, run shell commands for testing, and spawn subagents.
Sanitization: A dedicated 'Redaction Gate' in Phase 1 proactively scans all user input for sensitive patterns (API keys, tokens, private keys) and replaces them with environment variable placeholders.
[COMMAND_EXECUTION]: The skill automates the setup of a project dashboard by executing shell scripts for filesystem management (find, cp, ln) and process control (ps, kill). It also launches a local HTTP server using python3 -m http.server bound strictly to 127.0.0.1. These operations are benign and scoped to the skill's management directory.
[CREDENTIALS_SAFE]: The skill implements a robust redaction gate in phases/1-manifest.md that identifies credentials for Stripe, GitHub, AWS, Google, Slack, Telegram, and others, ensuring they are not saved in plain text or included in git history.

autopilot