The Agent Skills Directory

[SAFE]: The skill is designed with a strict 'no autonomous spend' policy. All outputs are markdown plans intended for human review and manual execution, with the 'gate_class' explicitly set to 'review' in the configuration.
[COMMAND_EXECUTION]: The skill utilizes local tools like Bash, Grep, and a project-specific Bun script (scripts/bootstrap-experience.ts) to manage project state and analyze performance artifacts. The script incorporates security best practices, such as symbolic link validation to prevent file system traversal attacks during project initialization.
[PROMPT_INJECTION]: The skill is architected to handle untrusted data from marketing research and performance logs, addressing indirect prompt injection risks through its multi-agent design.
Ingestion points: Reads from .forsvn/performance/*.tsv, campaign-plan.md, icp-research.md, and product-context.md.
Boundary markers: The system uses a 7-dimension quantitative rubric and a dedicated 'Critic' agent to verify the logical consistency and data grounding of the final output.
Capability inventory: Uses Bash, Read, Grep, and Glob tools for local file analysis.
Sanitization: Employs a 'Constraint Checker' agent to enforce hard rules (min-viable floors, concentration caps, fit vetoes) and an 'Anti-Patterns' detection catalog to reject unsafe or fabricated proposals.

plan-budget