The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions specify the installation of the @concord-ai/concord-mcp package from the public npm registry. It also suggests using npx -y @concord-ai/concord-mcp@latest for manual registration, which downloads and executes code from a remote source at runtime.
[COMMAND_EXECUTION]: The installation and initialization workflow involves executing shell commands such as npm install -g @concord-ai/concord-mcp, concord install, and concord init. These commands install a global CLI, modify local configuration files (e.g., .mcp.json, CLAUDE.md, .cursor/rules/), and create a local SQLite database in the repository.
[INDIRECT_PROMPT_INJECTION]: The skill manages a shared work-state where agents record decisions, findings, and task summaries. This creates an attack surface where malicious instructions could be embedded in task data to influence other agents that process the workspace.
Ingestion points: Untrusted data enters the context through the update_task, offer_handoff, and claim_work tools defined in SKILL.md.
Boundary markers: The skill documentation does not provide specific delimiters or instructions for the agent to ignore or sanitize embedded instructions within the shared task memory.
Capability inventory: The skill has the capability to write to persistent storage (SQLite) and modify core agent instruction files during its setup phase.
Sanitization: No sanitization or validation mechanisms are described for the natural language content stored in task contexts or handoff packets.

concord-mcp-workspace