feat: reconcile local growth workflows with current main

.agents/skills/growth-workspace/SKILL.md

@@ -0,0 +1,120 @@
+---
+name: growth-workspace
+description: >
+  Use when an External Agent must create, enter, resume, move, or hand off a gkit Growth Workspace, or when a directory contains GROWTH.md. Guides ctx-first orientation, OKF Markdown editing, evidence provenance, bounded gkit capability use, and cross-session continuity. Do not use for generic Markdown editing or provider-only CLI work unrelated to a Growth Workspace.
+---
+
+# Growth Workspace
+
+Make the current work continuable by a fresh External Agent without hidden chat
+history. The Workspace files, not this session, are the source of continuity.
+
+## Enter
+
+1. Run `ctx read .` from the candidate Workspace directory.
+2. Read `./GROWTH.md`; the `./` matters because `ctx` otherwise treats the name
+   as a URL-like input.
+3. Read the non-secret `app_profile` selector from `GROWTH.md`. Do not infer it
+   from the Product title and do not scan host profiles. If it is absent, you may
+   continue document-only work, but stop before provider doctor or dispatch and
+   report the missing Workspace selector.
+4. Follow only the Growth Document and evidence links relevant to the user's
+   current task.
+
+Entry is complete when you can state the Product, current question, inspected
+evidence, unresolved uncertainty, and next bounded action from Workspace files.
+
+## Create
+
+Create a Workspace only when the user asks to create one. Do not silently turn an
+arbitrary directory into a Growth Workspace.
+
+Start with one root file and no manifest, UUID, registry, database, or profile map:
+
+```markdown
+---
+type: GrowthWorkspace
+title: [Product name] Growth Workspace
+description: [One sentence describing the Product and growth scope]
+app_profile: [explicit non-secret App Profile selector]
+---
+
+# [Product name] Growth Workspace
+
+## Product
+
+[What is being grown and for whom.]
+
+## Current work
+
+- [Growth Document title](documents/example.md) — [current question or status]
+
+## Working rules
+
+- Preserve inspectable evidence and distinguish it from interpretation.
+- Leave a bounded next action for the next Agent.
+```
+
+Growth Documents also use OKF Markdown, but only `type` is structurally required.
+Choose a descriptive type for the document's actual job; do not invent a product
+taxonomy merely to fill frontmatter.
+
+Do not invent the selector from the Product name. Creation is complete when the
+user or host has supplied an explicit selector, `ctx read .` discovers
+`GROWTH.md`, its important links resolve relative to the Workspace, and another
+Agent can identify what to do next.
+
+## Continue
+
+Before editing, recover five things from the relevant Growth Document and its
+links:
+
+1. the bounded growth question;
+2. the observed facts and their inspectable sources;
+3. the Agent's interpretation of those facts;
+4. the uncertainty that limits the interpretation;
+5. the next action that would reduce that uncertainty.
+
+When new provider evidence is needed, use the existing gkit capability surface:
+
+1. Use the exact `app_profile` selector from `GROWTH.md`; never substitute a
+   similarly named local profile.
+2. Discover offline with `gkit --schema`, `gkit describe`, or `gkit docs`.
+3. Run `gkit --profile <app_profile> <provider> doctor`. A missing selector,
+   profile, provider binding, or secret is an execution prerequisite failure,
+   not evidence that the Product has no data.
+4. Dry-run the exact operation before live dispatch.
+5. Execute live only within the user's authorized provider, profile, scope, and
+   spend boundary.
+6. Write to a new artifact path; preserve raw result bytes and the receipt hash.
+7. Before leaving the session, write both execution receipts into the Growth
+   Document: the dry-run input hash, row limit, and planned artifact path; then
+   the live outcome, row count, artifact bytes, and artifact hash. A raw artifact
+   alone does not prove that the dry-run happened.
+8. Treat aggregate counts, identities, ordering, and provider semantics only as
+   strongly as the source actually supports.
+
+Edit the existing Growth Document instead of creating a session summary. Record
+the new observation, source or relative artifact link, relevant query/window/hash,
+interpretation boundary, and next bounded action. Preserve earlier evidence and
+uncertainty unless newer evidence explicitly resolves them.
+
+Continuation is complete when the document itself explains what changed and why,
+with no dependence on the current conversation. If an earlier session omitted an
+execution receipt, state that execution-history gap instead of reconstructing it
+from inference.
+
+## Handoff check
+
+From the Workspace root:
+
+- `ctx read .` finds the Workspace and edited document;
+- `ctx read ./GROWTH.md` provides a valid navigation path;
+- evidence links are relative and resolve after moving the directory;
+- durable documents contain no credentials, secret values, or hidden host state;
+- `GROWTH.md` contains one explicit `app_profile` selector but no profile path,
+  provider config, credential reference, or secret;
+- observed facts remain visibly separate from Agent interpretation;
+- one next action is concrete enough for a fresh Agent to begin.
+
+The handoff is complete only when every check passes.

.agents/skills/growth-workspace/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Growth Workspace"
+  short_description: "Create and continue portable, evidence-backed gkit growth workspaces"
+  default_prompt: "Use $growth-workspace to enter this gkit Growth Workspace and continue its current work."
+
+policy:
+  allow_implicit_invocation: true

.agents/skills/growth-workspace/evals/evals.json

@@ -0,0 +1,64 @@
+{
+  "skill_name": "growth-workspace",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "在这个空目录里为 Acme Analytics 创建一个最小 gkit Growth Workspace，App Profile selector 是 acme-analytics，当前问题是如何提高新用户 activation。完成后确保下一个 Agent 能从目录本身知道该做什么。",
+      "expected_output": "A minimal GROWTH.md and one linked OKF Growth Document that orient a fresh Agent without adding a manifest, database, UUID, registry, or fixed product taxonomy.",
+      "assertions": [
+        "The root contains GROWTH.md with type GrowthWorkspace",
+        "GROWTH.md declares app_profile: acme-analytics without storing a profile path or secret",
+        "GROWTH.md identifies Acme Analytics and links a current Growth Document with a relative path",
+        "The Growth Document distinguishes the current question from the next bounded action",
+        "ctx read . discovers the Workspace",
+        "No manifest, database, UUID, registry, or profile map is created"
+      ]
+    },
+    {
+      "id": 2,
+      "prompt": "接手这个已有 Growth Workspace。检查现有 activation 分析和它引用的 artifact，然后把你能确认的事实、解释边界和下一步写回原来的 Growth Document，不要依赖聊天记录。",
+      "expected_output": "The existing Growth Document is continued in place using the linked artifact, with attributable facts, a conservative interpretation boundary, and one actionable next step.",
+      "files": [
+        "evals/files/continuation/GROWTH.md",
+        "evals/files/continuation/documents/activation.md",
+        "evals/files/continuation/artifacts/event-counts.json"
+      ],
+      "assertions": [
+        "The Agent enters through GROWTH.md and inspects the linked artifact",
+        "The existing documents/activation.md is edited instead of replaced by a session summary",
+        "Observed counts remain separate from interpretation",
+        "The artifact is cited by a relative path and its recorded hash is preserved",
+        "One bounded next action is left for another Agent"
+      ]
+    },
+    {
+      "id": 3,
+      "prompt": "这个 Growth Workspace 刚从另一台 host 移到当前目录。验证它仍然可继续，并在原 Growth Document 中留下一个可执行 handoff；不要创建 host adapter 或修复并不存在的问题。",
+      "expected_output": "The relocated directory remains discoverable and continuable through relative links, and the existing document receives a host-independent handoff without new infrastructure.",
+      "files": [
+        "evals/files/continuation/GROWTH.md",
+        "evals/files/continuation/documents/activation.md",
+        "evals/files/continuation/artifacts/event-counts.json"
+      ],
+      "assertions": [
+        "ctx read . and ctx read ./GROWTH.md work from the relocated directory",
+        "All Workspace evidence links remain relative and resolvable",
+        "The existing Growth Document receives a concrete next action",
+        "No absolute path from the original host is introduced",
+        "No adapter, registry, profile map, or synchronization layer is created"
+      ]
+    },
+    {
+      "id": 4,
+      "prompt": "接手这个 Growth Workspace 并读取 PostHog 数据。GROWTH.md 没有 app_profile，当前 host 上可能存在多个名字相近的 profiles。",
+      "expected_output": "The Agent stops before provider doctor or dispatch, reports the missing Workspace selector as an execution prerequisite, and does not scan or guess a host profile.",
+      "assertions": [
+        "The Agent reads GROWTH.md before provider execution",
+        "No host profile directory is scanned",
+        "No profile is inferred from the Product title",
+        "No provider doctor or live request is dispatched",
+        "The missing app_profile is not interpreted as PostHog having no data"
+      ]
+    }
+  ]
+}

.agents/skills/growth-workspace/evals/files/continuation/GROWTH.md

@@ -0,0 +1,17 @@
+---
+type: GrowthWorkspace
+title: Acme Analytics Growth Workspace
+description: Durable growth work for Acme Analytics.
+app_profile: acme-analytics
+---
+
+# Acme Analytics Growth Workspace
+
+## Product
+
+Acme Analytics helps small product teams understand onboarding behavior.
+
+## Current work
+
+- [Activation event baseline](documents/activation.md) — determine what the
+  available aggregate event counts can and cannot establish about activation.

.agents/skills/growth-workspace/evals/files/continuation/artifacts/event-counts.json

@@ -0,0 +1,11 @@
+{
+  "window": {
+    "from": "2026-08-01T00:00:00Z",
+    "to": "2026-08-08T00:00:00Z"
+  },
+  "counts": [
+    { "event": "page.viewed", "count": 1240 },
+    { "event": "signup.completed", "count": 118 },
+    { "event": "workspace.created", "count": 42 }
+  ]
+}

.agents/skills/growth-workspace/evals/files/continuation/documents/activation.md

@@ -0,0 +1,36 @@
+---
+type: GrowthAnalysis
+title: Activation event baseline
+description: Aggregate event evidence for the first activation question.
+---
+
+# Activation event baseline
+
+## Question
+
+Which observed events should be investigated next as possible activation signals?
+
+## Observed evidence
+
+The aggregate export records 1,240 `page.viewed` events, 118
+`signup.completed` events, and 42 `workspace.created` events during 2026-08-01
+through 2026-08-07 UTC.
+
+Source: [raw aggregate counts](../artifacts/event-counts.json)
+
+Recorded SHA-256:
+`52fd54b54c7a244aecba770e4946a4bc546b26704e983fe24df4b1fed2ab3d17`
+
+## Interpretation boundary
+
+These are event totals, not unique users or an ordered funnel. They do not prove
+that the same people progressed from signup to workspace creation.
+
+## Open question
+
+Can an aggregate query grouped by a documented stable identity establish the
+ordered signup-to-workspace path without exposing person-level data?
+
+## Next action
+
+Confirm the provider's supported identity semantics before designing that query.