| name | knowledge-research |
| description | This skill should be used when the user asks to "search knowledge", "what do we know about X?", "look up prior decisions on X", "find conventions for X", or "research the knowledge base". |
Knowledge Research
Dispatch the engy:research subagent to search the workspace knowledge graph and return a curated digest of relevant findings.
Process
Step 1: Resolve Context
Identify scope hints from the current session:
- Active project name or slug
- Active milestone ref (e.g., "m7")
- Repo path if the question is repo-local
- workspaceId — required by
engy:research. If not already known, call listWorkspaces (or getWorkspaceDetails) to obtain it before dispatching.
Step 2: Dispatch Research Subagent
Invoke the engy:research subagent via the Task tool:
Task({
subagent_type: 'engy:research',
prompt: '<user question> — context: <active project/milestone/repo> workspaceId=<id>'
})
workspaceId is required — engy:research errors without it. Resolve it in Step 1 before dispatching.
Include any scope hints in the prompt so the subagent can apply filters.repo, filters.scenarioIds, or collection scoping when relevant.
Step 3: Present Results
Print the returned digest to the user verbatim. Do not summarize or reformat — the subagent's output is already structured for direct consumption.
Output Format
The subagent returns a ## Findings block with 3–8 cited findings and a sources/findings count footer. Present it as-is.
If the user's question is ambiguous, ask one clarifying question before dispatching (e.g., "Is this about the current project or workspace-wide?").
Key Principles
- Dispatch-only — this skill does no work beyond invoking the subagent.
- Present the digest verbatim — the subagent's structured output is already formatted for the user.
- Ask one clarifying question if the user's input is ambiguous (e.g., "is this workspace-wide or project-scoped?").
Flow Position
Typical trigger: user asks "what do we know about X?" in the terminal.
Next step: user uses the digest to inform their next action — write a plan, design a feature, file a bug.
