Run any Skill in Manus with one click

growthub-governed-mutation-loop

Stars47

Forks12

UpdatedJuly 8, 2026 at 23:32

Operate a governed Growthub workspace as an agent through the canonical loop — read state → reason with causal tools → dry-run preflight → governed hand-off through a sanctioned route → re-read. Use for ANY workspace-config change, sandbox/swarm execution, workflow publish, or helper apply inside a governed workspace or exported fork. This skill is how mutations happen; it never invents a third mutation path.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

Growthub-ai

Growthub-ai/growthub-local

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Computer Occupations, All OtherComputer and Mathematical Occupations·SOC 15-1299

SKILL.md

readonly

name	growthub-governed-mutation-loop
description	Operate a governed Growthub workspace as an agent through the canonical loop — read state → reason with causal tools → dry-run preflight → governed hand-off through a sanctioned route → re-read. Use for ANY workspace-config change, sandbox/swarm execution, workflow publish, or helper apply inside a governed workspace or exported fork. This skill is how mutations happen; it never invents a third mutation path.
triggers	["change the workspace config","patch the workspace","run a sandbox / swarm","publish a workflow","apply helper proposals","governed mutation","preflight this change"]
helpers	[]
subSkills	[{"name":"governed-workspace-mutation","path":"../oss-investigative-architecture/skills/governed-workspace-mutation/SKILL.md"}]
selfEval	{"criteria":["Every mutation routed through one of the four governed lanes (PATCH /api/workspace on an allowlisted key, POST /api/workspace/sandbox-run, POST /api/workspace/workflow/publish, POST /api/workspace/helper/apply) — no direct file writes, no invented routes.","POST /api/workspace/patch/preflight (or MCP preflight_patch) ran before every PATCH, and its verdict is quoted in the work log.","The success envelope of each governed call was confirmed by a re-read (GET /api/workspace or the relevant intelligence route) before any dependent step.","Rejections (400 allowed[], 422 violations[], AppScopeViolation) were treated as navigation — the repairPlan was followed, never bypassed.","No secret value appeared in any patch body, prompt, receipt, or logged response."],"maxRetries":3}

growthub-governed-mutation-loop

Contract docs (read before first mutation in a session): docs/OPERATING_THE_GOVERNED_UNIVERSE_V1.md (three-layer control plane) and docs/GOVERNED_MCP_CONSOLE_V1.md (agent console). The runtime-verified request/response card is the sub-skill governed-workspace-mutation (ships inside every exported workspace at skills/governed-workspace-mutation/SKILL.md) — load it before the first call.

The loop

READ → REASON → DRY-RUN → GOVERNED HAND-OFF → RE-READ

READ — GET /api/workspace (or MCP describe_workspace / get_workspace_topology). In --live MCP mode every call rehydrates; never act on remembered state.
REASON — use the causal tools before shaping the change: simulate_causal_impact, find_downstream_dependencies, trace_lineage (full recipe: growthub-causal-impact-analysis).
DRY-RUN — POST /api/workspace/patch/preflight (or MCP preflight_patch, which proxies it in --live and forwards x-growthub-app-scope). It uses the write path's exact merge step (applyWorkspaceConfigPatch), so it can never disagree with the real PATCH. A red preflight is an answer, not an obstacle.
GOVERNED HAND-OFF — emit only sanctioned routes:
- PATCH /api/workspace — ONE allowlisted key per patch (dashboards | widgetTypes | canvas | dataModel)
- POST /api/workspace/sandbox-run — all sandbox / agent-swarm execution (draft proofs via useDraft: true)
- POST /api/workspace/workflow/publish — the only draft → live transition (server-verified draftSha256 lineage)
- POST /api/workspace/helper/apply — human-reviewed proposals only MCP next_actions emits the exact governed call; MCP itself never mutates.
RE-READ — confirm the success envelope, then re-read state (and the receipt via growthub-outcome-receipts-bootstrap) before any dependent step.

Hard rules

The 4-key PATCH allowlist is the permanent ceiling; unknown top-level keys get 400 + allowed[]. Policy violations get 422 + structured violations[] with repairPlan[] — follow the plan, never work around the policy (apps/workspace/lib/workspace-patch-policy.js is runtime-enforced).
Live workflow fields (orchestrationGraph, version bumps, lifecycleStatus: "live") are publish-owned. Never PATCH them.
Read-modify-one-key: read the full config, modify exactly one allowlisted key, PATCH only that key.
Every governed action receipts to workspace:agent-outcomes — including blocked attempts. That is a feature; do not retry blindly to "clean" the ledger.
Prefer existing governed objects (scheduled jobs, api-registry rows, sandbox-environment rows, data views) over new code.
Secrets are env-ref names on every surface; credential-shaped values in a patch body are a 422 by design.

More from this repository

same repository

governed-console

Growthub-ai/growthub-local

Operate the governed-universe MCP tools over a Growthub workspace — describe the workspace, traverse the metadata graph, simulate causal impact and blast radius, check ship-readiness, dry-run a PATCH with preflight_patch, and convert findings into governed next actions. Use whenever you need to answer "what exists, what depends on what, what would break, is it ready, and what governed call should happen next" in a workspace that contains growthub.config.json.

2026-07-1947

workspace-causal-cli

Growthub-ai/growthub-local

Run the offline causal-intelligence CLI over a governed Growthub workspace — growthub plan (blast radius, stale surfaces, lineage), growthub patch (offline allowlisted edit with impact + compliance, PR-ready), growthub capture (live vs repo drift), growthub readiness (ship verdict across forks). Use for scripted or CI-side impact analysis when you want JSON output instead of MCP tool calls.

2026-07-1947

growthub-custom-workspace-starter

Growthub-ai/growthub-local

Baseline primitive for every Growthub governed workspace. Routing menu for the Custom Workspace Operator — delegates to the operator runbook (`skills.md`), the session-memory template (`templates/project.md`), the self-evaluation contract (`templates/self-eval.md`), the safe-shell helpers layer (`helpers/`), and the parallel sub-skill convention (`skills/`). Invoked when the user starts a greenfield workspace via `growthub starter init`, imports a GitHub repo or skills.sh skill via `growthub starter import-repo`/`import-skill`, or downloads this kit directly via `growthub kit download`.

2026-07-1947

governed-workspace-mutation

Growthub-ai/growthub-local

The two canonical Growthub workspace mutation calls and the verified protocol — PATCH /api/workspace (allowlist dashboards, widgetTypes, canvas, dataModel) and POST /api/workspace/sandbox-run. Read before changing workspace configuration or executing anything in a governed workspace. Mutations are runtime-enforced; this card keeps you on the sanctioned lanes.

2026-07-0847

workspace-helper

Growthub-ai/growthub-local

Draft dashboards, widgets, API registry rows, custom business objects, and swarm graphs for a governed Growthub workspace via the workspace helper — a propose-only planning engine whose output is applied through an explicit governed apply step. Use when asked to build a dashboard, create a custom object, register an API, scaffold a widget, repair broken references, or explain a workspace object.

2026-07-0847

growthub-auth

Growthub-ai/growthub-local

Sign in, confirm identity, and sign out of hosted Growthub via `growthub auth login` / `whoami` / `logout`. Use when the user needs to authenticate before Workflows / Saved Workflows / Pipeline execution, verify the active session, script auth with a pre-issued token, or opt out of telemetry.

2026-07-0847

name	growthub-governed-mutation-loop
description	Operate a governed Growthub workspace as an agent through the canonical loop — read state → reason with causal tools → dry-run preflight → governed hand-off through a sanctioned route → re-read. Use for ANY workspace-config change, sandbox/swarm execution, workflow publish, or helper apply inside a governed workspace or exported fork. This skill is how mutations happen; it never invents a third mutation path.
triggers	["change the workspace config","patch the workspace","run a sandbox / swarm","publish a workflow","apply helper proposals","governed mutation","preflight this change"]
helpers	[]
subSkills	[{"name":"governed-workspace-mutation","path":"../oss-investigative-architecture/skills/governed-workspace-mutation/SKILL.md"}]
selfEval	{"criteria":["Every mutation routed through one of the four governed lanes (PATCH /api/workspace on an allowlisted key, POST /api/workspace/sandbox-run, POST /api/workspace/workflow/publish, POST /api/workspace/helper/apply) — no direct file writes, no invented routes.","POST /api/workspace/patch/preflight (or MCP preflight_patch) ran before every PATCH, and its verdict is quoted in the work log.","The success envelope of each governed call was confirmed by a re-read (GET /api/workspace or the relevant intelligence route) before any dependent step.","Rejections (400 allowed[], 422 violations[], AppScopeViolation) were treated as navigation — the repairPlan was followed, never bypassed.","No secret value appeared in any patch body, prompt, receipt, or logged response."],"maxRetries":3}

growthub-governed-mutation-loop

The loop

READ → REASON → DRY-RUN → GOVERNED HAND-OFF → RE-READ

READ — GET /api/workspace (or MCP describe_workspace / get_workspace_topology). In --live MCP mode every call rehydrates; never act on remembered state.
REASON — use the causal tools before shaping the change: simulate_causal_impact, find_downstream_dependencies, trace_lineage (full recipe: growthub-causal-impact-analysis).
DRY-RUN — POST /api/workspace/patch/preflight (or MCP preflight_patch, which proxies it in --live and forwards x-growthub-app-scope). It uses the write path's exact merge step (applyWorkspaceConfigPatch), so it can never disagree with the real PATCH. A red preflight is an answer, not an obstacle.
GOVERNED HAND-OFF — emit only sanctioned routes:
- PATCH /api/workspace — ONE allowlisted key per patch (dashboards | widgetTypes | canvas | dataModel)
- POST /api/workspace/sandbox-run — all sandbox / agent-swarm execution (draft proofs via useDraft: true)
- POST /api/workspace/workflow/publish — the only draft → live transition (server-verified draftSha256 lineage)
- POST /api/workspace/helper/apply — human-reviewed proposals only MCP next_actions emits the exact governed call; MCP itself never mutates.
RE-READ — confirm the success envelope, then re-read state (and the receipt via growthub-outcome-receipts-bootstrap) before any dependent step.

Hard rules

The 4-key PATCH allowlist is the permanent ceiling; unknown top-level keys get 400 + allowed[]. Policy violations get 422 + structured violations[] with repairPlan[] — follow the plan, never work around the policy (apps/workspace/lib/workspace-patch-policy.js is runtime-enforced).
Live workflow fields (orchestrationGraph, version bumps, lifecycleStatus: "live") are publish-owned. Never PATCH them.
Read-modify-one-key: read the full config, modify exactly one allowlisted key, PATCH only that key.
Every governed action receipts to workspace:agent-outcomes — including blocked attempts. That is a feature; do not retry blindly to "clean" the ledger.
Prefer existing governed objects (scheduled jobs, api-registry rows, sandbox-environment rows, data views) over new code.
Secrets are env-ref names on every surface; credential-shaped values in a patch body are a 422 by design.