agentsclimarketplace

Growthub outcome receipts bootstrap

Skill Growthub-ai/growthub-local/.claude/skills/growthub-outcome-receipts-bootstrap

Start every governed-workspace session from the agent-outcome receipt ledger and self-correct from structured rejections. Use at the start of any session in a governed workspace or exported fork, after any 400/422/ AppScope rejection, when resuming interrupted work, or when auditing what agents actually did (blocked attempts, publishes, drafts awaiting proof). Receipts are the workspace's memory — read them before acting.From its SKILL.md

Install
npx -y skills add Growthub-ai/growthub-local --skill growthub-outcome-receipts-bootstrap

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.

SKILL.md

4.1 KB, as published. Nobody here has run it

growthub-outcome-receipts-bootstrap

Contract: AGENTS.md §"Agent Outcome Loop V1". Receipt builder source (starter kit): apps/workspace/lib/workspace-outcome-receipts.js. Every mutation lane — direct PATCH, preflight rejections, sandbox runs, workflow publishes, helper applies — emits the same canonical, secret-redacted receipt into the workspace:agent-outcomes source-record stream.

Session bootstrap (do this before acting)

  1. GET /api/workspace/agent-outcomes (or MCP outcome_ledger) → the receipt stream PLUS the derived governance summary: blocked attempts, publishes, drafts awaiting test/publish, live rows without proof.
  2. Read the newest receipts for the surface you are about to touch. Each carries receiptId, lane, verdict, nextActions, rollbackRef.
  3. If a prior session left nextActions, continue from them. If it left a failed/blocked outcome, its repairPlan[] is your first step — the rejection already encodes the navigation.
  4. Only then move to growthub-causal-impact-analysis (reason) and growthub-governed-mutation-loop (act).

Rejections are navigation, not noise

  • 400 + allowed[] — you used a non-allowlisted top-level key. Reshape to one of dashboards | widgetTypes | canvas | dataModel.
  • 422 + violations[] + repairPlan[] — the mutation policy caught a full-config body, credential-shaped field, oversized row/node-config, history blob, or live-workflow field. Apply the repair plan literally; each violation code names its fix.
  • AppScopeViolation + repairPlan[] — the x-growthub-app-scope header disagrees with the target rows; preflight's appScopeVerdict mirrors it, so a scoped preflight would have shown this first.
  • Blocked outcomes are receipted too — that is the audit trail working. Never spam retries of an unchanged call; change the call per the plan, bounded by this skill's maxRetries.

Lane classification (report honestly)

untrusted-direct (direct PATCH) · execution-proof (sandbox-run) · server-authoritative (workflow/publish) · governed-proposal (helper/apply — privileged, human-reviewed). When summarizing activity for the user, name the lane; the same edit means different trust levels on different lanes.

Hard rules

  • Never write to the receipt stream directly; receipts are emitted by the lanes themselves.
  • Never treat an empty ledger as "no governance" — it means a fresh workspace; the loop still applies.
  • Receipts reflect what was true when written; re-read live state before acting on an old nextActions.

Keep looking

Skills are one crate of 615,528. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.