feat(agents): add agents content type, built and served alongside skills

.github/workflows/build.yml

@@ -46,7 +46,7 @@ jobs:
     - name: Lint env var naming conventions
       run: bash scripts/lint-env-naming.sh
 
-      # No individual skill should be larger than 1MB, and the bundle should be less than 5MB.
+      # No individual skill should be larger than 1MB, and the bundle should be less than 8MB.
     - name: Check bundle and skill sizes
       run: |
         set -e
@@ -61,8 +61,8 @@ jobs:
         bundle_bytes=$(stat -c%s "$BUNDLE" 2>/dev/null || stat -f%z "$BUNDLE")
         echo "skills-mcp-resources.zip size: ${bundle_bytes} bytes"
 
-        if [ "$bundle_bytes" -gt $((5 * 1024 * 1024)) ]; then
-          echo "::error::skills-mcp-resources.zip is larger than 5MB (current: ${bundle_bytes} bytes)"
+        if [ "$bundle_bytes" -gt $((8 * 1024 * 1024)) ]; then
+          echo "::error::skills-mcp-resources.zip is larger than 8MB (current: ${bundle_bytes} bytes)"
           exit 1
         fi
 

context/agents/README.md

@@ -0,0 +1,51 @@
+# Agent prompts
+
+One `<flow>/<type>.md` per orchestrator task — the WHAT of the task. The folder
+is the flow: the wizard's registry is scoped per flow, so audit or migration
+agent sets get sibling folders. The frontmatter carries the artifacts the
+executor configures the run with: the model, the mini-skills to load (the HOW),
+the tools the task may and may not use, and the tasks it depends on. `flow`
+repeats the folder name so the file stays self-describing on disk (the build
+rejects a mismatch), and one prompt per flow is marked `seed: true`: the
+planner that seeds the queue, not an enqueueable task type.
+
+`model` is a gateway model id, not a runner binding — the wizard's switchboard
+routes it, and its CLI/flag rungs may override it per run. `allowedTools` /
+`disallowedTools` use the wizard's tool vocabulary; the executing harness maps
+them to its native tool names.
+
+The body is intent only — what to do and what done looks like. The client
+injects the basics (project context, how to report, how to surface progress),
+so a prompt never restates them. Frontmatter stays flat: scalars and inline
+`[a, b]` arrays only, the wizard's parser is deliberately not a YAML engine.
+
+This README is documentation, not data: the build serves every other `.md`
+under a flow folder as an agent prompt.
+
+Each agent names its model per harness profile — `model_pi`/`effort_pi` for the
+gpt (pi) run and `model_sdk`/`effort_sdk` for the anthropic (sdk) run — so the
+same flow benchmarks either provider. The mapping is not 1:1: a light task can be
+luna on pi but sonnet on sdk. Effort is optional and overrides the model default.
+
+```markdown
+---
+type: example
+flow: my-flow
+model_pi: openai/gpt-5.6-luna
+effort_pi: low
+model_sdk: claude-haiku-4-5-20251001
+skills: []
+allowedTools: [Read, Glob, Grep]
+disallowedTools: [enqueue_task]
+dependsOn: []
+---
+
+## Goal
+
+What this task does, in plain prose.
+
+## How you know you succeeded
+
+Plain-text success criteria live here. State what done looks like, and what to
+do when the task cannot be completed.
+```

context/agents/integration-v2/build.md

@@ -0,0 +1,30 @@
+---
+type: build
+flow: integration-v2
+label: Build and review the integration
+model_pi: openai/gpt-5.6-terra
+effort_pi: medium
+model_sdk: claude-sonnet-4-6
+effort_sdk: high
+skills: [integration-v2-build]
+allowedTools: [Read, Edit, Glob, Grep, Bash]
+disallowedTools: [enqueue_task]
+dependsOn: [install, init, identify, error-tracking, capture]
+---
+
+## Goal
+
+Bring the integration together and review it: install the dependencies the earlier
+steps declared, verify the project builds and lints, then review every change for
+convention fit and unintended edits. Until now the steps only edited code and the
+manifest — this is where it installs, is checked, and is reviewed as a whole.
+
+## How you know you succeeded
+
+The install completes, the integration is in place, and the changes read like the
+rest of the codebase with nothing unrelated touched or mangled. If the build or
+lint fails only on pre-existing errors you did not introduce, that still counts as
+done — note the conflict and finish. Reserve a failed status for when your own
+changes break the build. Put a one-line summary of any conflict in your handoff's
+`conflict` field and the full detail in what you did; the user sees the one-liner
+in the outro and the detail in the report.

context/agents/integration-v2/capture.md

@@ -0,0 +1,25 @@
+---
+type: capture
+flow: integration-v2
+label: Capture events
+model_pi: openai/gpt-5.6-terra
+effort_pi: medium
+model_sdk: claude-sonnet-4-6
+effort_sdk: high
+skills: [integration-v2-capture]
+allowedTools: [Read, Write, Edit, Glob, Grep]
+disallowedTools: [enqueue_task]
+dependsOn: [install, init]
+---
+
+## Goal
+
+Decide which events are worth capturing in this app, then instrument them in the
+same pass — read each file once, choose the events, and add the capture calls
+while the file is already open.
+
+## How you know you succeeded
+
+The meaningful user actions across the app have capture calls that fire on the
+real action, not on page load, and `.posthog-wizard-cache/.posthog-events.json`
+lists the events you instrumented.

context/agents/integration-v2/dashboard.md

@@ -0,0 +1,23 @@
+---
+type: dashboard
+flow: integration-v2
+label: Create a starter dashboard
+model_pi: openai/gpt-5.6-terra
+effort_pi: medium
+model_sdk: claude-sonnet-4-6
+effort_sdk: high
+skills: [integration-v2-dashboard, integration-v2-insight, integration-v2-mcp]
+allowedTools: [Read, Glob, Grep]
+disallowedTools: [Write, Edit, Bash, enqueue_task]
+dependsOn: [build]
+---
+
+## Goal
+
+Create a starter PostHog dashboard with a few insights built on the events this
+integration instruments, using the PostHog MCP.
+
+## How you know you succeeded
+
+A dashboard exists with a handful of insights on the captured events, and you hand
+off its URL for the report to link.

context/agents/integration-v2/error-tracking.md

@@ -0,0 +1,31 @@
+---
+type: error-tracking
+flow: integration-v2
+label: Add error tracking
+model_pi: openai/gpt-5.6-luna
+effort_pi: low
+model_sdk: claude-sonnet-4-6
+effort_sdk: high
+skills: [integration-v2-error-tracking-step]
+allowedTools: [Read, Write, Edit, Glob, Grep]
+disallowedTools: [enqueue_task]
+dependsOn: [install, init]
+---
+
+## Goal
+
+Set up the framework's single global error boundary so uncaught errors reach
+PostHog. One place — the init or app entry — following the docs and the reference
+example, not manual capture calls sprinkled across files. The SDK is already
+installed and initialized (see the context from previous steps); build on that,
+do not re-check it.
+
+This is an instrument-only task. Do not install dependencies, run the build, run
+tests, or start the app — a later `build` step does all verification. Stay inside
+this project's directory and edit the one global handler; that is the whole job.
+
+## How you know you succeeded
+
+A global error handler forwards exceptions to PostHog. You did not install
+anything, run a build or tests, search outside the project, or read through the
+whole app or hand-wrap individual components or routes.

context/agents/integration-v2/identify.md

@@ -0,0 +1,23 @@
+---
+type: identify
+flow: integration-v2
+label: Wire user identification
+model_pi: openai/gpt-5.6-terra
+effort_pi: medium
+model_sdk: claude-sonnet-4-6
+effort_sdk: high
+skills: [integration-v2-identify]
+allowedTools: [Read, Edit, Glob, Grep]
+disallowedTools: [enqueue_task]
+dependsOn: [install, init]
+---
+
+## Goal
+
+Wire user identification: call PostHog identify wherever the app establishes who
+the user is, typically at login and signup.
+
+## How you know you succeeded
+
+An identify call fires at the point the user becomes known, with a stable
+distinct id. If the app has no auth or user concept, say so and stop.

context/agents/integration-v2/init.md

@@ -0,0 +1,24 @@
+---
+type: init
+flow: integration-v2
+label: Set up PostHog initialization
+model_pi: openai/gpt-5.6-luna
+effort_pi: low
+model_sdk: claude-haiku-4-5-20251001
+skills: [integration-v2-init]
+allowedTools: [Read, Write, Edit, Glob, Grep]
+disallowedTools: [enqueue_task]
+dependsOn: []
+---
+
+## Goal
+
+Initialize PostHog: create the framework's init point so the SDK is configured
+once and available across the app, set the PostHog environment variables through
+the wizard tools, and document those keys in `.env.example` for other developers.
+
+## How you know you succeeded
+
+The init file exists and the PostHog env keys are present. Keys live in the env
+file, never hardcoded in source, and `.env.example` lists the key names (with
+placeholder values) so the next developer knows what to set.

context/agents/integration-v2/install.md

@@ -0,0 +1,23 @@
+---
+type: install
+flow: integration-v2
+label: Add the PostHog SDK to the manifest
+model_pi: openai/gpt-5.6-luna
+effort_pi: low
+model_sdk: claude-haiku-4-5-20251001
+skills: [integration-v2-install]
+allowedTools: [Read, Edit, Glob, Grep]
+disallowedTools: [enqueue_task]
+dependsOn: []
+---
+
+## Goal
+
+Declare the PostHog SDK in the project's package manifest. Do not run the package
+manager and do not build — the build task installs and verifies everything at the
+end.
+
+## How you know you succeeded
+
+The SDK is listed in the manifest's dependencies at a sensible version. If it is
+already declared, leave it and say so.

context/agents/integration-v2/integrate-posthog.md

@@ -0,0 +1,34 @@
+---
+type: integrate-posthog
+flow: integration-v2
+seed: true
+model_pi: openai/gpt-5.6-terra
+effort_pi: medium
+model_sdk: claude-sonnet-4-6
+effort_sdk: high
+skills: []
+allowedTools: [Read, Glob, Grep]
+disallowedTools: [Write, Edit, Bash, complete_task]
+dependsOn: []
+---
+
+## Goal
+
+Plan a PostHog integration and seed the task queue with this graph:
+
+- `install` and `init`, independent of each other.
+- `identify`, `capture`, and `error-tracking`, each after `install` and `init`
+  and independent of one another, so they run in parallel. `capture` decides the
+  events and instruments them; `error-tracking` wires the single global error
+  boundary — it needs the SDK installed and initialized, not the events.
+- `build`, after `install`, `init`, `identify`, `capture`, and `error-tracking` —
+  it installs the dependencies and verifies the project builds, lints, and passes
+  its tests.
+- `dashboard`, after `build` — only once the integration is confirmed building,
+  linting, and testing cleanly.
+- `report`, after `dashboard` — it writes the setup report last.
+
+## How you know you succeeded
+
+Every task in the graph is queued with that dependency shape, the report last,
+and the first task runnable. Keep labels short — the action in a few words.

context/agents/integration-v2/report.md

@@ -0,0 +1,27 @@
+---
+type: report
+flow: integration-v2
+label: Write the setup report
+model_pi: openai/gpt-5.6-luna
+effort_pi: low
+model_sdk: claude-sonnet-4-6
+effort_sdk: high
+skills: [integration-v2-report, integration-v2-notebook, integration-v2-mcp]
+allowedTools: [Read, Write, Glob, Grep]
+disallowedTools: [enqueue_task]
+dependsOn: [dashboard]
+---
+
+## Goal
+
+Write the setup report summarizing what this integration did, drawing only on the
+run's queue log and event plan in `.posthog-wizard-cache/` (`queue.json` and
+`.posthog-events.json`), then mirror it into a shareable PostHog notebook.
+
+## How you know you succeeded
+
+`posthog-setup-report.md` exists at the project root: what was installed and
+initialized, the events captured, whether identify was wired or skipped, error
+tracking added, the dashboard link, any build conflict in full, and the next
+steps for the user. The report is also mirrored into a PostHog notebook whose URL
+is emitted with the `[NOTEBOOK_URL]` marker.

context/docs.yaml

@@ -8,6 +8,7 @@ docs:
     tags: [core, users, identity]
     urls:
       - https://posthog.com/docs/getting-started/identify-users.md
+      - https://posthog.com/docs/product-analytics/identity-resolution.md
 
   - id: cloudflare-workers
     display_name: Cloudflare Workers

context/skills/integration-v2/build/config.yaml

@@ -0,0 +1,11 @@
+type: docs-only
+template: description.md
+description: Install dependencies and verify the project builds
+tags: [orchestrator, build]
+variants:
+  - id: all
+    display_name: PostHog build step
+    tags: [orchestrator, build]
+    docs_urls: []
+cli:
+  role: internal