Hivemind Goals Agent Skill

Hivemind Goals (openclaw)

OpenClaw exposes purpose-built tools for goals + KPIs. Use them directly — do NOT try to write files via the host filesystem.

Tools

• hivemind_goal_add({ text }) — create a new goal. Returns goal_id (UUID). Status starts at opened.
• hivemind_kpi_add({ goal_id, kpi_id, target, unit, name? }) — add a KPI to an existing goal. Only call when the user explicitly asks for KPIs; do NOT auto-generate.
• hivemind_search({ query }) — search Hivemind shared memory (summaries + sessions). Use this when the user asks "what's already there" before creating a duplicate.
• hivemind_read({ path }) — read the full content of a specific Hivemind path.
• hivemind_index({}) — list everything in memory.

Workflow when the user expresses a goal

• (Optional) hivemind_search first to surface any existing related goal.
• hivemind_goal_add({ text: "<short description>" }) — capture the returned goal_id.
• ONLY if the user asks for KPIs: hivemind_kpi_add once per KPI with goal_id + kpi_id (short slug like k-prs) + target (positive int) + unit.
• Confirm to the user with the goal_id and that the goal is team-visible.

Capture a task for later (with resumable context)

When the user parks a tangential task mid-session — "save this for later", "remind me to …", "don't let me forget …", "let's do X later" — store enough context to resume cold later, not just a one-liner. Put the full package in the text of hivemind_goal_add:

hivemind_goal_add({ text:
  "Add rate-limiting to the webhook handler\n\n" +
  "Start here: add a per-IP token bucket on the handler entry path\n" +
  "Files: src/webhook/handler.ts:120-160, src/webhook/limits.ts\n" +
  "Branch: feat/webhook-hardening\n" +
  "Run: pnpm test webhook\n" +
  "Why: bursty clients hammer the endpoint; defer until retry-backoff lands" })

Line 1 is the label. Fill Start here / Files / Branch / Run / Why from the conversation; Start here: (the concrete first action) matters most. (OpenClaw's hivemind_goal_add has no provenance flag, so the row is tagged manual — that's fine; the context is what matters.)

Resume a parked task (automatic context transfer)

When the user says "let's work on that task / goal" or "pick up the <X> task":

• hivemind_search({ query: "<topic>" }) or hivemind_index({}) to locate the parked goal, then hivemind_read({ path: "memory/goal/<owner>/opened/<goal_id>.md" }) to pull the full context package back.
• Read it as your working context and begin from Start here: using the Files / Branch / Run lines — continue as if the context was never lost.

(Status-move tools aren't exposed on OpenClaw, so leave the goal where it is and just resume the work.)

What NOT to do

• Do NOT write files anywhere under ~/.deeplake/memory/. OpenClaw's runtime does not route filesystem writes to the Deeplake tables — only the hivemind_* tools above do.
• Do NOT call hivemind_kpi_add unsolicited. Wait for the user to ask.
• Do NOT use hivemind_search to *create* anything — it's read-only.