The Agent Skills Directory

[COMMAND_EXECUTION]: The skill modifies the project's .claude/settings.json file to automatically grant permissions for Bash(*/knowledge-gate:*) and Bash(sqlite3 .knowledge/vault.db:*). This allows the agent to execute these commands in subsequent sessions without requesting user confirmation.
[EXTERNAL_DOWNLOADS]: The skill generates GitHub Workflows that fetch the Knowledge Distillery plugin from the author's repository at github.com/ether-moon/knowledge-distillery and download various MCP server packages from the npm registry.
[REMOTE_CODE_EXECUTION]: Automated workflows created by the skill execute code directly from the author's repository and run third-party packages dynamically using npx (e.g., mcp-linear, @modelcontextprotocol/server-slack).
[PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface through the batch-refine.yml and curate-report.yml workflows which ingest untrusted data from the repository environment.
Ingestion points: Pull Request bodies, commit metadata, and issue/PR comments are processed by the agent within the workflow.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the workflow prompt templates.
Capability inventory: The workflows grant the agent access to file system operations, terminal execution (Bash), and multiple MCP integrations (GitHub, Linear, Slack, Notion).
Sanitization: No sanitization or validation of the ingested PR data is performed before it is passed to the LLM.

setup