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
npx -y skills add Growthub-ai/growthub-local --skill growthub-outcome-receipts-bootstrapAssembled 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)
GET /api/workspace/agent-outcomes(or MCPoutcome_ledger) → the receipt stream PLUS the derived governance summary: blocked attempts, publishes, drafts awaiting test/publish, live rows without proof.- Read the newest receipts for the surface you are about to touch. Each
carries
receiptId, lane, verdict,nextActions,rollbackRef. - If a prior session left
nextActions, continue from them. If it left a failed/blocked outcome, itsrepairPlan[]is your first step — the rejection already encodes the navigation. - Only then move to
growthub-causal-impact-analysis(reason) andgrowthub-governed-mutation-loop(act).
Rejections are navigation, not noise
- 400 +
allowed[]— you used a non-allowlisted top-level key. Reshape to one ofdashboards|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[]— thex-growthub-app-scopeheader disagrees with the target rows; preflight'sappScopeVerdictmirrors 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.