The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands and specialized CLI tools to gather project metadata.
It executes git for branch state, status, and commit logs.
It uses the GitHub CLI (gh) to verify the status of pull requests and issues.
It uses bun to run helper TypeScript scripts and python3 for a related session mining script (dig.py).
Standard utilities like find, ls, grep, sed, and xargs are used to navigate the project structure and vault directories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources that could theoretically be manipulated by an attacker to influence the agent's orientation.
Ingestion points: Reads from the project vault (ψ/), specifically handoffs, retrospectives, focus files, and tracks. It also reads the agent's local project session logs (.jsonl).
Capability inventory: The skill has the ability to read files, run git/gh commands, and execute local scripts.
Sanitization: The skill primarily uses these inputs to construct descriptive tables and timelines, and explicitly instructs the agent to verify claims against the live system (e.g., checking if a file exists before reporting it as 'pending').
[SAFE]: The skill follows established platform conventions, such as using the ~/.claude directory for configuration and the ψ/ directory for project documentation. It does not perform unauthorized network operations, exfiltrate data, or attempt to persist access. The 'Verify Before Reporting' rule is a security positive that prevents the agent from being misled by stale or untrusted documentation.

recap