The Agent Skills Directory

[SAFE]: The skill incorporates robust security practices, including mandatory end-of-day security sweeps that utilize gitleaks for secret detection and semgrep for static analysis.
[COMMAND_EXECUTION]: The workflow requires the execution of standard development and orchestration commands such as git worktree management, dependency auditing (npm audit, pip-audit), and testing (pytest, npm test). These operations are well-defined within the session lifecycle and aim to maintain environment integrity.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted data from PROJECT_CONTEXT.md and automated observer reports.
Ingestion points: Project context files and quality reports read during the 'Context Scan' phase (SKILL.md, reference/start-day-protocol.md).
Boundary markers: The workflow enforces a 'Feature Contract' as a boundary to define expected outcomes before code ingestion (reference/dual-team-architecture.md).
Capability inventory: The agent can perform shell command execution, git operations, and network requests to webhooks (reference/feature-development.md).
Sanitization: The 'Devil's Advocate' role and the dual-team structure (Builder + Observer) provide constant adversarial review of all generated content and ingested context to prevent accidental obedience to injected instructions.
[SAFE]: Cost management logic utilizes budget enforcement and model routing to ensure efficient use of AI resources without compromising security or quality, using cheaper models for routine monitoring and more capable models for architecture decisions.

workflow-orchestrator