The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for installing and running the openwolf CLI tool, which performs significant system interactions including project scanning and hook registration.
Core commands like openwolf init and openwolf scan modify the project's .claude/settings.json and manage local state files in the .wolf/ directory.
The openwolf dashboard command initiates a local web server on port 18791.
The openwolf designqc command utilizes puppeteer-core to capture screenshots of local development environments for design review.
[DYNAMIC_EXECUTION]: The middleware functions by generating and executing Node.js scripts at runtime to intercept Claude Code's tool calls.
openwolf init writes six JavaScript hook scripts (e.g., pre-read.js, post-write.js) to the .wolf/hooks/ directory.
These scripts are executed as subprocesses by the agent platform during the file read/write lifecycle.
[PERSISTENCE]: The tool includes features for maintaining long-running background processes and scheduled tasks.
The openwolf daemon command facilitates background operation, optionally utilizing the PM2 process manager.
A cron system is described for executing periodic maintenance tasks, such as anatomy rescans and AI memory reflections.
[INDIRECT_PROMPT_INJECTION]: The tool's design centers on injecting content from project-specific files into the agent's prompt context to optimize token usage.
Ingestion points: Data is read from .wolf/anatomy.md, .wolf/cerebrum.md, and .wolf/buglog.json during the SessionStart and PreToolUse events.
Boundary markers: The documentation does not explicitly detail the use of XML delimiters or specific 'ignore' instructions for the injected summaries.
Capability inventory: The skill's environment includes capabilities for file system modification, background process management, and browser-based screenshot capture.
Sanitization: There is no mention of sanitization processes for the markdown content before it is emitted into the active context.

openwolf