The Agent Skills Directory

[DATA_EXFILTRATION]: The skill workflow involves inventorying sensitive local configuration, including model sheets, account details, and host restrictions. It specifically references reading project-related metadata from the ~/.claude/projects/ directory.
[PERSISTENCE]: The skill is designed to manage and install 'startup hooks' and 'plugin registrations', which allow for automated behavior and command execution at the start of future agent sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes external configuration data and model sheets which serve as injection surfaces.
Ingestion points: Reads pstack skill catalog, references/provider-dispatch.md, adapters/README.md, and ~/.claude/projects/ metadata.
Boundary markers: No explicit delimiters or instructions are specified to prevent the agent from following directives embedded within the processed configuration files.
Capability inventory: The skill possesses the ability to modify configuration files, register new plugins, and execute shell scripts such as worktree-audit.sh and run-hook.cmd.
Sanitization: There is no evidence of sanitization or strict schema validation for the external configuration data before it is interpolated into the setup workflow.
[EXTERNAL_DOWNLOADS]: The skill identifies as a port that incorporates code from several external repositories, including github.com/ericlitman/open-pstack and github.com/cursor/plugins. References to the anthropics/claude-plugins-official repository are noted as originating from a trusted organization.

setup-pstack