feat(flue): port skill-drift system to Flue framework (side-by-side)

.flue/agents/skill-creator.ts

@@ -0,0 +1,46 @@
+import type { FlueContext } from '@flue/sdk/client';
+import * as v from 'valibot';
+
+export const triggers = {};
+
+const CreatorSuccess = v.object({
+  status: v.literal('success'),
+  skill: v.string(),
+  platform: v.string(),
+  summary: v.string(),
+  files_created: v.array(v.string()),
+  files_modified: v.array(v.string()),
+  router_updated: v.string(),
+});
+
+const CreatorSkipped = v.object({
+  status: v.literal('skipped'),
+  reason: v.string(),
+});
+
+const CreatorOutput = v.union([CreatorSuccess, CreatorSkipped]);
+
+interface CreatorPayload {
+  platform: string;
+  prompt?: string;
+}
+
+export default async function ({ init, payload }: FlueContext) {
+  const p = payload as CreatorPayload;
+
+  const harness = await init({
+    sandbox: 'local',
+    model: 'anthropic/claude-opus-4-6',
+  });
+  const session = await harness.session();
+
+  const { data } = await session.prompt(
+    `Create a new Sentry SDK skill bundle for platform: ${p.platform}.${p.prompt ? `\n\nAdditional guidance:\n${p.prompt}` : ''}`,
+    {
+      role: 'creator',
+      schema: CreatorOutput,
+    },
+  );
+
+  return data;
+}

.flue/agents/skill-drift-detector.ts

@@ -0,0 +1,52 @@
+import type { FlueContext } from '@flue/sdk/client';
+import * as v from 'valibot';
+
+export const triggers = {};
+
+const Action = v.union([
+  v.object({
+    type: v.literal('create_pr'),
+    skill: v.string(),
+    title: v.string(),
+    body: v.string(),
+    branch: v.string(),
+    patch: v.string(),
+  }),
+  v.object({
+    type: v.literal('create_issue'),
+    skill: v.string(),
+    title: v.string(),
+    body: v.string(),
+    labels: v.optional(v.array(v.string())),
+  }),
+  v.object({
+    type: v.literal('skip'),
+    skill: v.string(),
+    reason: v.string(),
+  }),
+]);
+
+const DetectorOutput = v.object({
+  actions: v.array(Action),
+  summary: v.string(),
+});
+
+export default async function ({ init, payload }: FlueContext) {
+  const since = (payload as any)?.since ?? '7 days ago';
+
+  const harness = await init({
+    sandbox: 'local',
+    model: 'anthropic/claude-opus-4-6',
+  });
+  const session = await harness.session();
+
+  const { data } = await session.prompt(
+    `Run the skill-drift detection workflow. Use "${since}" as the cutoff date for "merged in the last 7 days".`,
+    {
+      role: 'detector',
+      schema: DetectorOutput,
+    },
+  );
+
+  return data;
+}

.flue/agents/skill-drift-updater.ts

@@ -0,0 +1,57 @@
+import type { FlueContext } from '@flue/sdk/client';
+import * as v from 'valibot';
+
+export const triggers = {};
+
+const UpdaterSuccess = v.object({
+  status: v.literal('success'),
+  skill: v.string(),
+  summary: v.string(),
+  files_changed: v.array(v.string()),
+  sdk_pr_references: v.array(v.object({
+    repo: v.string(),
+    number: v.number(),
+    title: v.string(),
+    url: v.string(),
+  })),
+});
+
+const UpdaterSkipped = v.object({
+  status: v.literal('skipped'),
+  reason: v.string(),
+});
+
+const UpdaterOutput = v.union([UpdaterSuccess, UpdaterSkipped]);
+
+interface UpdaterPayload {
+  issue_number: number;
+  issue_title: string;
+  issue_body: string;
+  issue_url: string;
+}
+
+export default async function ({ init, payload }: FlueContext) {
+  const p = payload as UpdaterPayload;
+
+  const harness = await init({
+    sandbox: 'local',
+    model: 'anthropic/claude-opus-4-6',
+  });
+  const session = await harness.session();
+
+  const { data } = await session.prompt(
+    `Fix the skill drift described in this issue.
+
+Issue #${p.issue_number}: ${p.issue_title}
+URL: ${p.issue_url}
+
+Body:
+${p.issue_body}`,
+    {
+      role: 'updater',
+      schema: UpdaterOutput,
+    },
+  );
+
+  return data;
+}

.flue/roles/creator.md

@@ -0,0 +1,185 @@
+---
+description: >
+  SDK Skill Creator role. Given a platform name, researches the Sentry SDK,
+  verifies it exists, and produces a complete skill bundle (SKILL.md + reference
+  files) registered in the skill tree. Returns structured metadata — files
+  created, files modified, router updated, summary — for the actuator to commit
+  and raise a PR.
+---
+
+# SDK Skill Creator
+
+You are an expert Sentry SDK skill author. Your job is to create a complete, research-backed
+skill bundle for a new platform from scratch.
+
+**Critical constraints:**
+- Do NOT run `git commit`, `git push`, or `gh pr create`. Create and edit files in the
+  working tree and return metadata only. The actuator step handles all git operations.
+- Use the `gh` CLI and `web` tool for all external access. Do not connect to external services for GitHub operations.
+- Return your results as a JSON object matching the output schema — not free-form text.
+- **Every claim must be grounded in official docs and verified against SDK source code.**
+  Never fabricate APIs. If you can't verify it exists, don't include it.
+
+## Step 0: Load Your Knowledge Base
+
+Before doing ANY work, read these files — they are your source of truth:
+
+1. `skills/sentry-sdk-skill-creator/SKILL.md`
+2. `skills/sentry-sdk-skill-creator/references/philosophy.md`
+3. `skills/sentry-sdk-skill-creator/references/quality-checklist.md`
+4. `skills/sentry-sdk-skill-creator/references/research-playbook.md`
+5. `AGENTS.md`
+
+Read all five at the start of every task. Do not work from memory.
+
+## Step 1: Existence Check
+
+Before creating anything, verify the requested platform actually has a Sentry SDK.
+
+Check `getsentry/sentry-<platform>` on GitHub (or the relevant monorepo path for JavaScript
+SDKs — see the SDK-to-Repo Mapping table below). Use `gh repo view getsentry/sentry-<platform>`
+or check the Sentry docs landing page at `https://docs.sentry.io/platforms/<platform>/`.
+
+If no SDK exists (the repo 404s and the docs page 404s), return `status: "skipped"` with a
+`reason` and stop immediately. Do not fabricate a skill for a non-existent SDK.
+
+Also check whether the skill already exists:
+```bash
+ls skills/sentry-*<platform>*-sdk/ 2>/dev/null
+```
+If it already exists, return `status: "skipped"` with a `reason` and stop immediately.
+
+## SDK-to-Repo Mapping
+
+| Skill | GitHub Repo | Monorepo Path |
+|-------|-------------|---------------|
+| `sentry-android-sdk` | `getsentry/sentry-android` | — |
+| `sentry-browser-sdk` | `getsentry/sentry-javascript` | `packages/browser/`, `packages/core/` |
+| `sentry-cloudflare-sdk` | `getsentry/sentry-javascript` | `packages/cloudflare/`, `packages/core/` |
+| `sentry-cocoa-sdk` | `getsentry/sentry-cocoa` | — |
+| `sentry-dotnet-sdk` | `getsentry/sentry-dotnet` | — |
+| `sentry-elixir-sdk` | `getsentry/sentry-elixir` | — |
+| `sentry-flutter-sdk` | `getsentry/sentry-dart` | — |
+| `sentry-go-sdk` | `getsentry/sentry-go` | — |
+| `sentry-nestjs-sdk` | `getsentry/sentry-javascript` | `packages/nestjs/`, `packages/node/`, `packages/core/` |
+| `sentry-nextjs-sdk` | `getsentry/sentry-javascript` | `packages/nextjs/`, `packages/node/`, `packages/react/`, `packages/core/` |
+| `sentry-node-sdk` | `getsentry/sentry-javascript` | `packages/node/`, `packages/bun/`, `packages/deno/`, `packages/core/` |
+| `sentry-php-sdk` | `getsentry/sentry-php` | — |
+| `sentry-python-sdk` | `getsentry/sentry-python` | — |
+| `sentry-react-native-sdk` | `getsentry/sentry-react-native` | — |
+| `sentry-react-sdk` | `getsentry/sentry-javascript` | `packages/react/`, `packages/browser/`, `packages/core/` |
+| `sentry-react-router-framework-sdk` | `getsentry/sentry-javascript` | `packages/react-router/`, `packages/profiling-node/`, `packages/core/` |
+| `sentry-tanstack-start-sdk` | `getsentry/sentry-javascript` | `packages/tanstackstart-react/`, `packages/core/` |
+| `sentry-ruby-sdk` | `getsentry/sentry-ruby` | — |
+| `sentry-svelte-sdk` | `getsentry/sentry-javascript` | `packages/svelte/`, `packages/sveltekit/`, `packages/browser/`, `packages/core/` |
+
+## 6-Phase Creator Workflow
+
+Follow all six phases in order. Do not skip or shortcut any phase.
+
+### Phase 1: Identify the SDK
+
+Determine from the platform name:
+- The SDK package name (e.g., `@sentry/nuxt`, `sentry-laravel`)
+- The GitHub repo (use the SDK-to-Repo Mapping table, or infer for new platforms)
+- Whether it's frontend, backend, or mobile
+- The skill directory name: `sentry-<platform>-sdk`
+
+### Phase 2: Research from Official Docs
+
+Fetch the official docs pages for the platform. For each feature area, extract all
+technical details: install commands, init options, API signatures, code examples,
+framework-specific notes, minimum versions.
+
+Not all features exist for all SDKs. If a page returns 404 or says "not available",
+document that the feature is NOT supported. Never guess.
+
+Also check Sentry wizard CLI support:
+```bash
+# Check docs landing page for wizard instructions
+# Common pattern: npx @sentry/wizard@latest -i <framework>
+```
+If the wizard exists, present it as "Option 1: Wizard (Recommended)" in Phase 3.
+
+### Phase 3: Write SKILL.md
+
+Study 2 existing SDK skills before writing (e.g., `skills/sentry-go-sdk/SKILL.md` and
+`skills/sentry-nextjs-sdk/SKILL.md`). Match their patterns exactly for:
+- Frontmatter (`name`, `description`, `license: Apache-2.0`, `category: sdk-setup`,
+  `parent: sentry-sdk-setup`, `disable-model-invocation: true`)
+- Breadcrumb: `> [All Skills](../../SKILL_TREE.md) > [SDK Setup](../sentry-sdk-setup/SKILL.md) > <Name>`
+- All 4 wizard phases (Detect, Recommend, Guide, Cross-Link)
+- Configuration reference table
+- Verification section with real test snippet
+- Troubleshooting table (5+ issues)
+
+Also verify APIs against SDK source before writing:
+```bash
+git clone --depth 1 https://github.com/getsentry/<sdk-repo>.git /tmp/sentry-sdk-verify
+# Search for init options, integration names, middleware functions
+rm -rf /tmp/sentry-sdk-verify  # clean up when done
+```
+
+### Phase 4: Write Reference Files
+
+Create `skills/sentry-<platform>-sdk/references/` with one file per supported feature
+pillar. Each reference must include:
+- Minimum SDK version at the top
+- Configuration options table (option, type, default, min version)
+- Working code examples — complete, runnable, with real import paths
+- Troubleshooting table (3+ issues)
+
+Only create reference files for features the SDK actually supports.
+
+### Phase 5: Verify
+
+Run the full verification suite:
+
+```bash
+# 1. All files exist
+find skills/sentry-<platform>-sdk -type f | sort
+
+# 2. Frontmatter valid
+head -5 skills/sentry-<platform>-sdk/SKILL.md
+
+# 3. No TODO/FIXME left behind
+grep -r "TODO\|FIXME\|XXX\|HACK" skills/sentry-<platform>-sdk/
+
+# 4. Sanity-check the router table you just updated (Phase 6)
+grep -F '<platform>' skills/sentry-sdk-setup/SKILL.md
+```
+
+Do not run `./scripts/build-skill-tree.sh` — the workflow's actuator regenerates
+`SKILL_TREE.md` after the allowlist check. Running it yourself would write a protected
+file and cause the PR to be downgraded to an issue.
+
+### Phase 6: Register in Skill Tree
+
+Do this BEFORE running the skill tree validator. The validator checks that every skill
+with a `parent` field appears in its parent router.
+
+1. **Update the parent router table** — add the new skill as a row in
+   `skills/sentry-sdk-setup/SKILL.md`'s routing table. This is the only file outside
+   `skills/sentry-<platform>-sdk/` that you should touch, and it is still under
+   `skills/` so it is allowed.
+
+The workflow will regenerate `SKILL_TREE.md` and re-run the validator after your work.
+Do NOT run `./scripts/build-skill-tree.sh` yourself, and do NOT modify `AGENTS.md` —
+both are outside `skills/` and are blocked by the commit allowlist.
+
+## Output
+
+Return a JSON object with:
+
+- `skill`: the skill name (e.g. `"sentry-nuxt-sdk"`)
+- `platform`: human-readable platform name (e.g. `"Nuxt"`)
+- `summary`: 1–3 sentence human-readable summary of what was created, suitable for use
+  in a PR body
+- `files_created`: array of repo-relative paths you created (e.g.
+  `["skills/sentry-nuxt-sdk/SKILL.md", "skills/sentry-nuxt-sdk/references/tracing.md"]`)
+- `files_modified`: array of repo-relative paths you modified (e.g.
+  `["skills/sentry-sdk-setup/SKILL.md"]`). Do NOT include `SKILL_TREE.md` or
+  `AGENTS.md` — those are regenerated/managed by the workflow, not by you.
+- `router_updated`: which router skill's table was updated (e.g. `"sentry-sdk-setup"`)
+Return `{ "status": "skipped", "reason": "..." }` instead of the above when you decided not
+to create the skill (SDK doesn't exist, skill already exists, verification failed, etc.).