fix(embed): support hosted Perplexity embeddings (pplx-embed-v1-*) (#1046)

Adds a `perplexity` embedding recipe (OpenAI-compatible at
https://api.perplexity.ai/v1, auth via PERPLEXITY_API_KEY only — never an
OPENAI_API_KEY fallback) covering pplx-embed-v1-0.6b and pplx-embed-v1-4b.

Perplexity's /embeddings endpoint diverges from OpenAI's wire shape in two
places that break the AI SDK adapter, handled by a new perplexityCompatFetch
shim (mirrors the Voyage/ZeroEntropy pattern incl. the two-layer OOM caps):
- encoding_format only accepts base64_int8/base64_binary; the SDK's 'float'
  default is forced to 'base64_int8' outbound.
- The response embedding is base64-encoded signed int8 components (natively
  quantized); decoded to number[] inbound so the SDK's Zod schema validates.
  Cosine similarity is scale-invariant, so raw int8 components rank correctly.

Flexible dims (Matryoshka-style 128..native max: 1024 for 0.6b, 2560 for 4b)
validate fail-loud in dims.ts + the init preflight; `dimensions` is
Perplexity's native field so no wire translation is needed. default_dims is
1024 (works on a plain vector column for both models); the 4b model's full
2560 width rides the existing halfvec (>2000 dims) storage/ANN path. Pricing
entries land in embedding-pricing.ts.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

llms-full.txt

@@ -1411,8 +1411,7 @@ This is the dispatcher. Skills are the implementation. **Read the skill file bef
 | "get more out of gbrain", "is my brain set up right", "weekly brain checkup", "advise me on my brain", "gbrain advisor" | `skills/gbrain-advisor/SKILL.md` |
 | Save or load reports | `skills/reports/SKILL.md` |
 | "Create a skill", "improve this skill" | `skills/skill-creator/SKILL.md` |
-| "save this learning to the vault", "capture this skill in Obsidian", "record this workflow in my notes", "put this setup change in the vault" | `skills/skill-vault-capture-policy/SKILL.md` |
-| "Skillify this", "is this a skill?", "make this proper", "add tests and evals for this" | `skills/skillify/SKILL.md` |
+| "Skillify this", "is this a skill?", "make this proper" | `skills/skillify/SKILL.md` |
 | "Compress my resolver", "AGENTS.md too large", "RESOLVER.md too big", "functional area dispatcher", "shrink routing table" | `skills/functional-area-resolver/SKILL.md` |
 | "Is gbrain healthy?", morning health check, skillpack-check | `skills/skillpack-check/SKILL.md` |
 | "harvest this skill into gbrain", "publish this skill to gbrain", "lift this skill upstream", "share this skill with other gbrain clients", "promote my skill to gbrain" | `skills/skillpack-harvest/SKILL.md` |
@@ -1430,7 +1429,6 @@ This is the dispatcher. Skills are the implementation. **Read the skill file bef
 | "Set up GBrain", first boot | `skills/setup/SKILL.md` |
 | "Now what?", "fill my brain", "cold start", "bootstrap", "import my data", "what should I import first" | `skills/cold-start/SKILL.md` |
 | "Migrate from Obsidian/Notion/Logseq" | `skills/migrate/SKILL.md` |
-| "Connect Obsidian to gbrain", "import my vault to gbrain", "sync vault and gbrain", "embed gbrain after vault update", "is gbrain synced with my vault" | `skills/obsidian-gbrain-safe-index/SKILL.md` |
 | Brain health check, maintenance run | `skills/maintain/SKILL.md` |
 | "Extract links", "build link graph", "populate timeline" | `skills/maintain/SKILL.md` (extraction sections) |
 | "Run dream", "process today's session", "synthesize my conversations", "consolidate yesterday's conversations", "what patterns did you see", "did the dream cycle run" | `skills/maintain/SKILL.md` (dream cycle section) |

openclaw.plugin.json

@@ -39,8 +39,8 @@
     "skills/briefing",
     "skills/citation-fixer",
     "skills/concept-synthesis",
-    "skills/cron-scheduler",
     "skills/cross-modal-review",
+    "skills/cron-scheduler",
     "skills/daily-task-manager",
     "skills/daily-task-prep",
     "skills/data-research",
@@ -54,11 +54,10 @@
     "skills/media-ingest",
     "skills/meeting-ingestion",
     "skills/minion-orchestrator",
-    "skills/obsidian-gbrain-safe-index",
     "skills/perplexity-research",
     "skills/query",
-    "skills/repo-architecture",
     "skills/reports",
+    "skills/repo-architecture",
     "skills/signal-detector",
     "skills/skill-creator",
     "skills/skillify",

skills/RESOLVER.md

@@ -60,8 +60,7 @@ This is the dispatcher. Skills are the implementation. **Read the skill file bef
 | "get more out of gbrain", "is my brain set up right", "weekly brain checkup", "advise me on my brain", "gbrain advisor" | `skills/gbrain-advisor/SKILL.md` |
 | Save or load reports | `skills/reports/SKILL.md` |
 | "Create a skill", "improve this skill" | `skills/skill-creator/SKILL.md` |
-| "save this learning to the vault", "capture this skill in Obsidian", "record this workflow in my notes", "put this setup change in the vault" | `skills/skill-vault-capture-policy/SKILL.md` |
-| "Skillify this", "is this a skill?", "make this proper", "add tests and evals for this" | `skills/skillify/SKILL.md` |
+| "Skillify this", "is this a skill?", "make this proper" | `skills/skillify/SKILL.md` |
 | "Compress my resolver", "AGENTS.md too large", "RESOLVER.md too big", "functional area dispatcher", "shrink routing table" | `skills/functional-area-resolver/SKILL.md` |
 | "Is gbrain healthy?", morning health check, skillpack-check | `skills/skillpack-check/SKILL.md` |
 | "harvest this skill into gbrain", "publish this skill to gbrain", "lift this skill upstream", "share this skill with other gbrain clients", "promote my skill to gbrain" | `skills/skillpack-harvest/SKILL.md` |
@@ -79,7 +78,6 @@ This is the dispatcher. Skills are the implementation. **Read the skill file bef
 | "Set up GBrain", first boot | `skills/setup/SKILL.md` |
 | "Now what?", "fill my brain", "cold start", "bootstrap", "import my data", "what should I import first" | `skills/cold-start/SKILL.md` |
 | "Migrate from Obsidian/Notion/Logseq" | `skills/migrate/SKILL.md` |
-| "Connect Obsidian to gbrain", "import my vault to gbrain", "sync vault and gbrain", "embed gbrain after vault update", "is gbrain synced with my vault" | `skills/obsidian-gbrain-safe-index/SKILL.md` |
 | Brain health check, maintenance run | `skills/maintain/SKILL.md` |
 | "Extract links", "build link graph", "populate timeline" | `skills/maintain/SKILL.md` (extraction sections) |
 | "Run dream", "process today's session", "synthesize my conversations", "consolidate yesterday's conversations", "what patterns did you see", "did the dream cycle run" | `skills/maintain/SKILL.md` (dream cycle section) |

skills/manifest.json

@@ -2,7 +2,7 @@
   "name": "gbrain",
   "version": "0.32.3.0",
   "conformance_version": "1.0.0",
-  "description": "Personal knowledge brain with hybrid RAG search — GStack mod for agent platforms",
+  "description": "Personal knowledge brain with hybrid RAG search \u2014 GStack mod for agent platforms",
   "skills": [
     {
       "name": "ingest",
@@ -34,11 +34,6 @@
       "path": "migrate/SKILL.md",
       "description": "Universal migration from Obsidian, Notion, Logseq, markdown, CSV, JSON, Roam"
     },
-    {
-      "name": "obsidian-gbrain-safe-index",
-      "path": "obsidian-gbrain-safe-index/SKILL.md",
-      "description": "Connect and sync an Obsidian-style Markdown vault with gbrain using a cost-controlled import-first workflow, explicit embedding gates, and durable skill/workflow capture."
-    },
     {
       "name": "setup",
       "path": "setup/SKILL.md",
@@ -268,11 +263,6 @@
       "name": "skill-optimizer",
       "path": "skill-optimizer/SKILL.md",
       "description": "Self-evolving skill optimization via gbrain skillopt — SkillOpt-paper-grounded text-space optimizer with validation gating (median-of-3 + epsilon=0.05), bundled-skill safety, bootstrap review sentinel, per-skill DB lock, and atomic versioned writes."
-    },
-    {
-      "name": "skill-vault-capture-policy",
-      "path": "skill-vault-capture-policy/SKILL.md",
-      "description": "Capture durable operational learnings, new agent skills, and environment/setup changes into the Obsidian vault instead of transient chat memory."
     }
   ],
   "dependencies": {

skills/obsidian-gbrain-safe-index/SKILL.md

@@ -1,137 +0,0 @@
----
-name: obsidian-gbrain-safe-index
-version: 1.0.0
-description: |
-  Connect, maintain, and sync an Obsidian-style Markdown vault with gbrain while preserving a cost-controlled workflow: the vault remains the source of truth, gbrain is the searchable/embedded index, durable skills/workflows are captured into the vault, and paid embedding runs only after explicit approval.
-triggers:
-  - "connect Obsidian to gbrain"
-  - "import my vault to gbrain"
-  - "sync vault and gbrain"
-  - "capture this skill in my vault"
-  - "embed gbrain after vault update"
-  - "is gbrain synced with my vault"
-tools:
-  - terminal
-  - read_file
-  - search_files
-  - write_file
-  - patch
-mutating: true
----
-
-# Obsidian → gbrain Safe Index and Capture
-
-## Contract
-
-This skill guarantees:
-
-- Treats the user's Obsidian-style Markdown vault as the source of truth before gbrain indexing.
-- Keeps gbrain in conservative mode unless the user explicitly approves a more expensive mode.
-- Imports vault changes with `--no-embed` first, then embeds only after explicit approval for the specific paid action.
-- Captures durable new skills, workflows, and environment learnings into the vault instead of leaving them only in chat memory.
-- Verifies every sync with concrete `gbrain stats`, search mode, and, when embeddings run, exact embedded chunk counts.
-
-## Phases
-
-1. **Resolve the vault path.**
-   - Prefer an existing environment variable such as `OBSIDIAN_VAULT_PATH` or `WIKI_PATH`.
-   - If no path is configured, search likely note directories and ask the user before writing.
-   - Verify the directory exists and contains markdown files or an `.obsidian` directory.
-
-2. **Read vault operating rules before writing.**
-   - If the vault has `SCHEMA.md`, `index.md`, `log.md`, `AGENTS.md`, or similar operating files, read them before ingest/query/major edit.
-   - Respect immutable source folders such as `raw/` when the vault declares them.
-   - Use the vault's native link convention, usually Obsidian `[[wikilinks]]`, for durable relationships.
-
-3. **MECE/capture decision.**
-   - If new knowledge belongs on an existing page, update that page.
-   - If it is a distinct recurring workflow or operational policy, create a small meta or concept page following the vault schema.
-   - Update the vault index/catalog for every new page when the vault maintains one.
-   - Append a log entry for meaningful vault updates when the vault maintains a log.
-
-4. **Safe gbrain import path.**
-   - Pre-check source directory; do not import a nonexistent path.
-   - Run `gbrain config set search.mode conservative` before/after risky reinit steps.
-   - Run `gbrain import "$OBSIDIAN_VAULT_PATH" --no-embed`.
-   - Run `gbrain extract links --source fs --dir "$OBSIDIAN_VAULT_PATH"` when wikilinks changed materially.
-
-5. **Paid embedding gate.**
-   - Do not run `gbrain embed --stale` unless the user explicitly asks or a prior instruction clearly approved this exact paid action.
-   - Before embedding, verify provider readiness with `gbrain providers test --model <provider:model>`.
-   - Confirm the configured embedding dimensions match the local schema.
-   - After embedding, verify `gbrain stats` and record exact `Pages`, `Chunks`, `Embedded`, and `Links` counts.
-
-6. **Final verification and vault echo.**
-   - Run `gbrain stats` and `gbrain search modes`.
-   - If vault files changed, re-import with `--no-embed`; if embedding was approved, embed stale chunks afterward.
-   - Report what changed, what was free/local, what used API billing, and what remains pending.
-
-## Output Format
-
-Use a compact status table:
-
-| Item | Status |
-|---|---|
-| Vault path | `/path` |
-| Vault updated | yes/no + files |
-| gbrain mode | conservative/balanced/tokenmax |
-| Import | `--no-embed` completed / skipped / failed |
-| Pages/chunks | exact counts from `gbrain stats` |
-| Embeddings | exact count; note whether this run used API billing |
-| Links | exact count |
-| Background jobs | none / list exact jobs |
-
-Then include:
-
-- **Safe next step:** free/local action.
-- **Paid next step:** embedding/LLM action, if any, with explicit approval requirement.
-
-## Anti-Patterns
-
-- Creating a duplicate skill/page when an existing Obsidian, gbrain, or vault-ingest skill already covers the workflow.
-- Running `gbrain embed --stale`, `gbrain dream`, `gbrain autopilot --install`, `gbrain onboard --auto`, or `tokenmax` without explicit cost approval.
-- Importing a nonexistent or wrong directory and treating a zero-page import as success.
-- Forgetting to update the vault index/catalog and log after creating or materially updating vault pages.
-- Recording API keys, tokens, or raw secrets in the vault or final response.
-
-## Tools Used
-
-- `read_file` — read vault schema/index/log and target notes.
-- `search_files` — find existing vault pages and avoid duplicates.
-- `write_file` / `patch` — create or update vault pages.
-- `terminal` — run `gbrain`, `git`, and environment checks with secret values redacted.
-
-## Safe Commands
-
-```bash
-export PATH="$HOME/.bun/bin:$PATH"
-gbrain config set search.mode conservative
-gbrain import "$OBSIDIAN_VAULT_PATH" --no-embed
-gbrain extract links --source fs --dir "$OBSIDIAN_VAULT_PATH"
-gbrain stats
-gbrain search modes
-```
-
-## Paid / Approval-Gated Commands
-
-```bash
-gbrain providers test --model <provider:model>
-gbrain embed --stale
-gbrain dream
-gbrain autopilot --install
-gbrain onboard --auto --max-usd 5
-gbrain config set search.mode tokenmax
-```
-
-## Verification Checklist
-
-- [ ] Vault path exists and is the intended source.
-- [ ] Vault operating files were read before edits when present.
-- [ ] Existing pages/skills were searched to avoid duplicates.
-- [ ] New/updated vault pages follow the vault schema and link convention.
-- [ ] Vault index/catalog updated for new pages when present.
-- [ ] Vault log appended for meaningful actions when present.
-- [ ] `gbrain import ... --no-embed` completed.
-- [ ] `gbrain stats` recorded pages/chunks/embeddings/links.
-- [ ] `gbrain search modes` confirms conservative mode unless a different mode was explicitly approved.
-- [ ] No paid/background commands ran without approval.

skills/obsidian-gbrain-safe-index/routing-eval.jsonl

@@ -1,11 +0,0 @@
-// Routing eval fixtures for skills/obsidian-gbrain-safe-index.
-// Positive cases: intents embed a trigger phrase in natural surrounding context
-// (never verbatim-identical to a trigger — the fixture linter rejects tautologies).
-{"intent": "help me connect Obsidian to gbrain for my notes", "expected_skill": "obsidian-gbrain-safe-index"}
-{"intent": "import my vault to gbrain but skip embeddings for now", "expected_skill": "obsidian-gbrain-safe-index"}
-{"intent": "please sync vault and gbrain after I edit notes", "expected_skill": "obsidian-gbrain-safe-index"}
-{"intent": "run embed gbrain after vault update tonight", "expected_skill": "obsidian-gbrain-safe-index"}
-{"intent": "hey is gbrain synced with my vault right now", "expected_skill": "obsidian-gbrain-safe-index"}
-// Negative cases: related but owned by other skills. Assert NO route to this skill.
-{"intent": "migrate my notes from Notion to gbrain", "expected_skill": null}
-{"intent": "what is on my calendar tomorrow", "expected_skill": null}

skills/skill-vault-capture-policy/SKILL.md

@@ -1,100 +0,0 @@
----
-name: skill-vault-capture-policy
-version: 1.0.0
-description: |
-  Use when the user wants a durable operational learning, new agent skill, or
-  important environment/setup change to be captured into the Obsidian vault
-  instead of left only in transient chat memory. Covers the capture rule,
-  preferred page patterns, and index/log update obligations.
-triggers:
-  - "save this learning to the vault"
-  - "capture this skill in Obsidian"
-  - "record this workflow in my notes"
-  - "put this setup change in the vault"
-  - "should we add this to the knowledge base"
-tools:
-  - read_file
-  - search_files
-  - write_file
-  - patch
-mutating: true
----
-
-# Skill Vault Capture Policy
-
-## Contract
-
-This skill guarantees:
-
-- Durable operational learnings, newly adopted skills, and important environment/setup changes are captured into the Obsidian vault rather than left only in chat memory.
-- An existing page is updated when the knowledge clearly belongs there; a new page is created only when the topic is distinct and likely to recur.
-- Every new vault page is added to `index.md`.
-- Every meaningful create/update appends a dated entry to `log.md`.
-- Small linked pages are preferred over one giant running note.
-
-## Phases
-
-1. **Classify the learning.**
-   - New gbrain operating rule, cost control, or embedding/provider change.
-   - New agent-fork / harness / coding-tool integration fact.
-   - New Obsidian vault workflow or structure decision.
-   - New recurring agent skill that changes how the agent should operate here.
-
-2. **Avoid duplicates.**
-   - Search the vault for an existing page that already owns the topic.
-   - If found, update it with a new section or dated note rather than creating a near-duplicate.
-
-3. **Create when distinct.**
-   - Place new pages under the vault schema: `_meta/` for operating notes, `concepts/` for workflows, `entities/` for tools/people.
-   - Use YAML frontmatter and at least two `[[wikilinks]]` unless it is a short seed page.
-
-4. **Update navigation.**
-   - Add the page to `index.md` under the correct type heading.
-   - Append a `## [YYYY-MM-DD] create|update | subject` entry to `log.md`.
-
-5. **Report the capture.**
-   - State which files changed and whether `index.md` / `log.md` were updated.
-
-## Output Format
-
-Use a short status block:
-
-| Item | Status |
-|---|---|
-| Learning classified | type |
-| Page created/updated | path |
-| index.md updated | yes/no |
-| log.md updated | yes/no |
-
-## Anti-Patterns
-
-- Leaving durable learnings only in chat memory.
-- Creating a near-duplicate page instead of updating the existing one.
-- Forgetting to update `index.md` and `log.md`.
-- Writing one giant running note instead of small linked pages.
-- Recording secrets, API keys, or raw credentials in the vault.
-
-## Tools Used
-
-- `read_file` — read `SCHEMA.md`, `index.md`, `log.md`, and target pages.
-- `search_files` — find existing pages to avoid duplicates.
-- `write_file` / `patch` — create or update vault pages and navigation.
-
-## Safe Commands
-
-```bash
-# inspect vault navigation before writing
-read SCHEMA.md index.md log.md
-# create or update a page, then refresh catalog/log
-# index.md: add [[page-slug]] under the matching type heading
-# log.md: append ## [YYYY-MM-DD] create|update | subject
-```
-
-## Verification Checklist
-
-- [ ] Vault schema/read files were checked before writing.
-- [ ] Existing pages were searched to avoid duplicates.
-- [ ] New/updated page has frontmatter and wikilinks.
-- [ ] `index.md` updated for new pages.
-- [ ] `log.md` appended for meaningful actions.
-- [ ] No secrets or raw credentials were written.

skills/skill-vault-capture-policy/routing-eval.jsonl

@@ -1,10 +0,0 @@
-// Routing eval fixtures for skills/skill-vault-capture-policy.
-// Positive cases: intents embed a trigger phrase in natural surrounding context
-// (never verbatim-identical to a trigger — the fixture linter rejects tautologies).
-{"intent": "please save this learning to the vault so we keep it", "expected_skill": "skill-vault-capture-policy", "ambiguous_with": ["idea-ingest"]}
-{"intent": "we should capture this skill in Obsidian for reuse", "expected_skill": "skill-vault-capture-policy", "ambiguous_with": ["capture"]}
-{"intent": "can you record this workflow in my notes for next time", "expected_skill": "skill-vault-capture-policy"}
-{"intent": "put this setup change in the vault before we forget", "expected_skill": "skill-vault-capture-policy"}
-// Negative cases: related but owned by other skills or out of scope.
-{"intent": "connect my Obsidian vault to gbrain", "expected_skill": null}
-{"intent": "what is on my calendar tomorrow", "expected_skill": null}

src/core/ai/dims.ts

@@ -90,6 +90,30 @@ export function isValidOpenAITextEmbedding3Dim(modelId: string, dims: number): b
   return Number.isInteger(dims) && dims >= 1 && dims <= max;
 }
 
+// Perplexity hosted embeddings (#1046): Matryoshka-style flexible dims,
+// any integer from 128 up to the model's native size. `dimensions` is the
+// native wire field (no translation needed); output encoding divergence
+// (base64 int8) is handled by perplexityCompatFetch in gateway.ts.
+const PERPLEXITY_EMBEDDING_MAX_DIMS: Record<string, number> = {
+  'pplx-embed-v1-0.6b': 1024,
+  'pplx-embed-v1-4b': 2560,
+};
+export const PERPLEXITY_MIN_DIMS = 128;
+
+export function isPerplexityEmbeddingModel(modelId: string): boolean {
+  return modelId in PERPLEXITY_EMBEDDING_MAX_DIMS;
+}
+
+export function maxPerplexityEmbeddingDim(modelId: string): number | undefined {
+  return PERPLEXITY_EMBEDDING_MAX_DIMS[modelId];
+}
+
+export function isValidPerplexityDim(modelId: string, dims: number): boolean {
+  const max = PERPLEXITY_EMBEDDING_MAX_DIMS[modelId];
+  if (max === undefined) return false;
+  return Number.isInteger(dims) && dims >= PERPLEXITY_MIN_DIMS && dims <= max;
+}
+
 // NVIDIA NIM hosted embedding models use asymmetric input_type values. Most
 // emit fixed natural dimensions, but llama-nemotron-embed-1b-v2 accepts
 // Matryoshka-style dimension overrides (e.g. matching an existing 1280d
@@ -226,6 +250,23 @@ export function dimsProviderOptions(
           },
         };
       }
+      // Perplexity pplx-embed-v1-* — flexible dims via the native
+      // `dimensions` field. Fail-loud when the configured dim is outside
+      // the model's range (same rationale as the Voyage/ZE guards: the
+      // upstream HTTP 400 misroutes as a transient network error).
+      // Symmetric retrieval — inputType is never emitted.
+      if (isPerplexityEmbeddingModel(modelId)) {
+        if (!isValidPerplexityDim(modelId, dims)) {
+          const max = maxPerplexityEmbeddingDim(modelId)!;
+          throw new AIConfigError(
+            `Perplexity model "${modelId}" supports embedding_dimensions in ` +
+            `${PERPLEXITY_MIN_DIMS}..${max}, got ${dims}.`,
+            `Set \`embedding_dimensions\` to a value between ${PERPLEXITY_MIN_DIMS} and ${max} ` +
+            `in your gbrain config.`,
+          );
+        }
+        return { openaiCompatible: { dimensions: dims } };
+      }
       // NVIDIA NIM hosted embeddings are OpenAI-compatible but require
       // asymmetric input_type. Use passage for indexing/document-side vectors
       // and query for search-side vectors. Only llama-nemotron-embed-1b-v2

src/core/ai/gateway.ts

@@ -263,6 +263,18 @@ export class ZeroEntropyResponseTooLargeError extends Error {
   }
 }
 
+/** Perplexity twin of the Voyage/ZE OOM caps (#1046). Int8 components are
+ * 1 byte each, so a real response (512 texts × 2560 dims) is ~1.3 MB —
+ * anything near this cap is unambiguously not legitimate. */
+const MAX_PERPLEXITY_RESPONSE_BYTES = 256 * 1024 * 1024;
+
+export class PerplexityResponseTooLargeError extends Error {
+  constructor(message: string) {
+    super(message);
+    this.name = 'PerplexityResponseTooLargeError';
+  }
+}
+
 // ---- Unified auth resolution (D12=A) ----
 //
 // Pre-v0.32, openai-compatible auth was duplicated across instantiateEmbedding,
@@ -1292,6 +1304,103 @@ const openAICompatAsymmetricFetch = (async (input: RequestInfo | URL, init?: Req
   return fetch(typeof input === 'string' ? input : input.toString(), baseInit);
 }) as unknown as typeof fetch;
 
+/**
+ * Perplexity compatibility shim (#1046). Perplexity's `/v1/embeddings`
+ * endpoint is OpenAI-shaped but diverges on two points that break the AI
+ * SDK's openai-compatible adapter:
+ *   - `encoding_format` only accepts 'base64_int8' (default) or
+ *     'base64_binary'; the SDK sends 'float', which Perplexity rejects.
+ *     Force 'base64_int8' on the wire.
+ *   - The response `embedding` is a base64 string encoding SIGNED INT8
+ *     components (natively quantized output). The SDK schema expects
+ *     `number[]` — decode Int8Array → number[] here. Cosine similarity is
+ *     scale-invariant, so the raw int8 components rank correctly.
+ * `dimensions` is Perplexity's native field name — no translation needed
+ * (dims.ts emits it directly). Layer 1/Layer 2 OOM caps mirror the Voyage
+ * pattern.
+ *
+ * Exported for tests (behavioral coverage of the int8 decode); not part of
+ * the public gateway API.
+ */
+export const perplexityCompatFetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
+  // OUTBOUND: force the encoding Perplexity actually accepts.
+  if (init?.body && typeof init.body === 'string') {
+    try {
+      const parsed = JSON.parse(init.body);
+      if (parsed && typeof parsed === 'object' && parsed.encoding_format !== 'base64_int8') {
+        parsed.encoding_format = 'base64_int8';
+        // Drop Content-Length so fetch recomputes from the new body.
+        const headers = new Headers(init.headers ?? {});
+        headers.delete('content-length');
+        init = { ...init, body: JSON.stringify(parsed), headers };
+      }
+    } catch {
+      // Body wasn't JSON — pass through untouched.
+    }
+  }
+
+  const resp = await fetch(input as any, init);
+  if (!resp.ok) return resp;
+  const ct = resp.headers.get('content-type') ?? '';
+  if (!ct.toLowerCase().includes('application/json')) return resp;
+
+  // Layer 1: Content-Length pre-check BEFORE the body is parsed.
+  const contentLengthHeader = resp.headers.get('content-length');
+  if (contentLengthHeader) {
+    const len = parseInt(contentLengthHeader, 10);
+    if (Number.isFinite(len) && len > MAX_PERPLEXITY_RESPONSE_BYTES) {
+      throw new PerplexityResponseTooLargeError(
+        `Perplexity response Content-Length=${len} exceeds ${MAX_PERPLEXITY_RESPONSE_BYTES} bytes — ` +
+        `likely compromised endpoint or misconfiguration`,
+      );
+    }
+  }
+
+  // INBOUND: decode base64 int8 embeddings to number[] so the SDK's Zod
+  // schema validates.
+  try {
+    const json: any = await resp.clone().json();
+    if (!json || typeof json !== 'object') return resp;
+    let modified = false;
+    if (Array.isArray(json.data)) {
+      for (const item of json.data) {
+        if (item && typeof item.embedding === 'string') {
+          // Layer 2: per-embedding cap for chunked responses that skipped
+          // Layer 1. base64 → bytes is the canonical 0.75 ratio.
+          const estDecoded = Math.ceil(item.embedding.length * 0.75);
+          if (estDecoded > MAX_PERPLEXITY_RESPONSE_BYTES) {
+            throw new PerplexityResponseTooLargeError(
+              `Perplexity embedding base64 exceeds ${MAX_PERPLEXITY_RESPONSE_BYTES} bytes ` +
+              `(estimated ${estDecoded} bytes from ${item.embedding.length} base64 chars)`,
+            );
+          }
+          // base64_int8: one signed int8 per component.
+          const bytes = Buffer.from(item.embedding, 'base64');
+          item.embedding = Array.from(new Int8Array(bytes.buffer, bytes.byteOffset, bytes.byteLength));
+          modified = true;
+        }
+      }
+    }
+    if (json.usage && typeof json.usage === 'object' && json.usage.prompt_tokens === undefined) {
+      json.usage.prompt_tokens = typeof json.usage.total_tokens === 'number'
+        ? json.usage.total_tokens
+        : 0;
+      modified = true;
+    }
+    if (!modified) return resp;
+    return new Response(JSON.stringify(json), {
+      status: resp.status,
+      statusText: resp.statusText,
+      headers: resp.headers,
+    });
+  } catch (err) {
+    // OOM-cap throws MUST propagate; anything else falls back to the
+    // original response (same contract as voyageCompatFetch).
+    if (err instanceof PerplexityResponseTooLargeError) throw err;
+    return resp;
+  }
+}) as unknown as typeof fetch;
+
 async function resolveEmbeddingProvider(modelStr: string): Promise<{ model: any; recipe: Recipe; modelId: string }> {
   const { parsed, recipe } = resolveRecipe(modelStr);
   assertTouchpoint(recipe, 'embedding', parsed.modelId, getExtendedModelsForProvider(parsed.providerId));
@@ -1360,6 +1469,8 @@ function instantiateEmbedding(recipe: Recipe, modelId: string, cfg: AIGatewayCon
           ? zeroEntropyCompatFetch
           : recipe.id === 'nvidia'
           ? nvidiaCompatFetch
+          : recipe.id === 'perplexity'
+          ? perplexityCompatFetch
           : openAICompatAsymmetricFetch);
       const client = createOpenAICompatible({
         name: recipe.id,

src/core/ai/recipes/index.ts

@@ -26,6 +26,7 @@ import { llamaServerReranker } from './llama-server-reranker.ts';
 import { moonshot } from './moonshot.ts';
 import { mistral } from './mistral.ts';
 import { nvidia } from './nvidia.ts';
+import { perplexity } from './perplexity.ts';
 
 const ALL: Recipe[] = [
   openai,
@@ -48,6 +49,7 @@ const ALL: Recipe[] = [
   moonshot,
   mistral,
   nvidia,
+  perplexity,
 ];
 
 /** Map from `provider:id` key to recipe. */

src/core/ai/recipes/perplexity.ts

@@ -0,0 +1,54 @@
+import type { Recipe } from '../types.ts';
+
+/**
+ * Perplexity's hosted embeddings API (#1046). OpenAI-shaped at
+ * `POST {base}/embeddings` but diverges on the wire:
+ *   - `encoding_format` only accepts 'base64_int8' (default) or
+ *     'base64_binary' — the AI SDK's 'float' default is rejected.
+ *   - The response `embedding` is a base64 string encoding SIGNED INT8
+ *     components (natively quantized output), not a float array.
+ * Both divergences are handled by perplexityCompatFetch in gateway.ts
+ * (force 'base64_int8' outbound; decode Int8Array → number[] inbound).
+ * Cosine similarity is scale-invariant, so the raw int8 components store
+ * and rank correctly as floats.
+ *
+ * Models (per docs.perplexity.ai/api-reference/embeddings-post, 2026-07):
+ *   - pplx-embed-v1-0.6b: dims 128..1024 (default 1024)
+ *   - pplx-embed-v1-4b:   dims 128..2560 (default 2560)
+ * The flexible-dim range validation lives in src/core/ai/dims.ts
+ * (PERPLEXITY_EMBEDDING_MAX_DIMS). default_dims is pinned at 1024 so both
+ * models work out of the box on a plain vector(N) column; users who want
+ * the 4b model's full 2560 width set `embedding_dimensions: 2560` and the
+ * existing halfvec path (dims > 2000) covers storage + ANN.
+ *
+ * Auth is PERPLEXITY_API_KEY only — deliberately NO OPENAI_API_KEY
+ * fallback (a Perplexity brain must never silently bill/route through
+ * OpenAI). If your key lives in PPLX_API_KEY, re-export it.
+ */
+export const perplexity: Recipe = {
+  id: 'perplexity',
+  name: 'Perplexity',
+  tier: 'openai-compat',
+  implementation: 'openai-compatible',
+  base_url_default: 'https://api.perplexity.ai/v1',
+  auth_env: {
+    required: ['PERPLEXITY_API_KEY'],
+    setup_url: 'https://www.perplexity.ai/settings/api',
+  },
+  touchpoints: {
+    embedding: {
+      models: ['pplx-embed-v1-0.6b', 'pplx-embed-v1-4b'],
+      default_dims: 1024,
+      cost_per_1m_tokens_usd: 0.03, // pplx-embed-v1-4b; 0.6b is $0.004/M
+      price_last_verified: '2026-07-21',
+      // Perplexity enforces 120K combined tokens (and 512 texts) per
+      // request. Same pre-split posture as Voyage: assume a dense
+      // tokenizer (1 char ≈ 1 token) at 0.5 utilization; the gateway's
+      // recursive halving is the runtime safety net.
+      max_batch_tokens: 120_000,
+      chars_per_token: 1,
+      safety_factor: 0.5,
+    },
+  },
+  setup_hint: 'Get an API key at https://www.perplexity.ai/settings/api, then `export PERPLEXITY_API_KEY=...` (re-export PPLX_API_KEY if that is where your key lives).',
+};

src/core/embedding-dim-check.ts

@@ -32,6 +32,10 @@ import {
   nvidiaEmbeddingDim,
   nvidiaEmbeddingDimOptions,
   supportsNvidiaEmbeddingDimension,
+  isPerplexityEmbeddingModel,
+  isValidPerplexityDim,
+  maxPerplexityEmbeddingDim,
+  PERPLEXITY_MIN_DIMS,
 } from './ai/dims.ts';
 
 /**
@@ -462,6 +466,15 @@ function isCustomDimValidForProvider(
         `(allowed: ${ZEROENTROPY_VALID_DIMS.join(', ')}).`,
     };
   }
+  if (recipe.id === 'perplexity' && isPerplexityEmbeddingModel(modelId)) {
+    if (isValidPerplexityDim(modelId, requestedDims)) return { valid: true, error: '' };
+    return {
+      valid: false,
+      error:
+        `Perplexity ${modelId} accepts dimensions ${PERPLEXITY_MIN_DIMS}..${maxPerplexityEmbeddingDim(modelId)}, ` +
+        `got ${requestedDims}.`,
+    };
+  }
   if (recipe.id === 'openai' && isOpenAITextEmbedding3Model(modelId)) {
     if (isValidOpenAITextEmbedding3Dim(modelId, requestedDims)) return { valid: true, error: '' };
     const maxDim = maxOpenAITextEmbedding3Dim(modelId);

src/core/embedding-pricing.ts

@@ -40,6 +40,9 @@ export const EMBEDDING_PRICING: Record<string, EmbeddingPricing> = {
   // Mistral (https://mistral.ai/pricing/api/, verified 2026-07-19)
   'mistral:mistral-embed':         { pricePerMTok: 0.10 },
   'mistral:mistral-embed-2312':    { pricePerMTok: 0.10 },
+  // Perplexity (https://docs.perplexity.ai/getting-started/pricing, verified 2026-07-21)
+  'perplexity:pplx-embed-v1-0.6b': { pricePerMTok: 0.004 },
+  'perplexity:pplx-embed-v1-4b':   { pricePerMTok: 0.03 },
 };
 
 export type PriceLookupResult =

test/ai/recipe-perplexity.test.ts

@@ -0,0 +1,142 @@
+/**
+ * #1046 — Perplexity hosted embeddings (pplx-embed-v1-*).
+ *
+ * Covers the three seams the recipe touches:
+ *  - recipe registration + auth (PERPLEXITY_API_KEY only, never OPENAI_API_KEY)
+ *  - flexible-dim validation (128..native max) in dims.ts + the init
+ *    preflight (resolveSchemaEmbeddingDim), incl. the >2000-dim 4b case
+ *  - perplexityCompatFetch: forces encoding_format=base64_int8 outbound and
+ *    decodes the base64 int8 embedding payload to number[] inbound
+ */
+import { afterEach, describe, expect, test } from 'bun:test';
+import {
+  dimsProviderOptions,
+  isPerplexityEmbeddingModel,
+  isValidPerplexityDim,
+  maxPerplexityEmbeddingDim,
+} from '../../src/core/ai/dims.ts';
+import { getRecipe, RECIPES } from '../../src/core/ai/recipes/index.ts';
+import { perplexity } from '../../src/core/ai/recipes/perplexity.ts';
+import { defaultResolveAuth, perplexityCompatFetch } from '../../src/core/ai/gateway.ts';
+import { AIConfigError } from '../../src/core/ai/errors.ts';
+import { resolveSchemaEmbeddingDim } from '../../src/core/embedding-dim-check.ts';
+import { lookupEmbeddingPrice } from '../../src/core/embedding-pricing.ts';
+
+describe('recipe: perplexity', () => {
+  test('registered as an OpenAI-compatible embedding provider', () => {
+    expect(RECIPES.has('perplexity')).toBe(true);
+    expect(getRecipe('perplexity')).toBe(perplexity);
+    expect(perplexity.tier).toBe('openai-compat');
+    expect(perplexity.implementation).toBe('openai-compatible');
+    expect(perplexity.base_url_default).toBe('https://api.perplexity.ai/v1');
+    const e = perplexity.touchpoints.embedding!;
+    expect(e.models).toEqual(['pplx-embed-v1-0.6b', 'pplx-embed-v1-4b']);
+    expect(e.default_dims).toBe(1024);
+    expect(e.max_batch_tokens).toBe(120_000);
+  });
+
+  test('auth is PERPLEXITY_API_KEY bearer — no OPENAI_API_KEY fallback', () => {
+    expect(perplexity.resolveAuth).toBeUndefined();
+    expect(perplexity.auth_env?.required).toEqual(['PERPLEXITY_API_KEY']);
+    expect(defaultResolveAuth(perplexity, { PERPLEXITY_API_KEY: 'fake-pplx' }, 'embedding')).toEqual({
+      headerName: 'Authorization',
+      token: 'Bearer fake-pplx',
+    });
+    // An OPENAI_API_KEY in the env must NOT satisfy Perplexity auth.
+    expect(() => defaultResolveAuth(perplexity, { OPENAI_API_KEY: 'sk-test' }, 'embedding')).toThrow(AIConfigError);
+  });
+
+  test('dims: 128..native-max range per model', () => {
+    expect(isPerplexityEmbeddingModel('pplx-embed-v1-4b')).toBe(true);
+    expect(maxPerplexityEmbeddingDim('pplx-embed-v1-4b')).toBe(2560);
+    expect(maxPerplexityEmbeddingDim('pplx-embed-v1-0.6b')).toBe(1024);
+    expect(isValidPerplexityDim('pplx-embed-v1-4b', 2560)).toBe(true);
+    expect(isValidPerplexityDim('pplx-embed-v1-4b', 128)).toBe(true);
+    expect(isValidPerplexityDim('pplx-embed-v1-4b', 64)).toBe(false);
+    expect(isValidPerplexityDim('pplx-embed-v1-0.6b', 2560)).toBe(false);
+  });
+
+  test('dimsProviderOptions emits native `dimensions`, fails loud out of range', () => {
+    expect(dimsProviderOptions('openai-compatible', 'pplx-embed-v1-4b', 2560)).toEqual({
+      openaiCompatible: { dimensions: 2560 },
+    });
+    // Symmetric provider — inputType never emitted.
+    expect(dimsProviderOptions('openai-compatible', 'pplx-embed-v1-4b', 1024, 'query')).toEqual({
+      openaiCompatible: { dimensions: 1024 },
+    });
+    expect(() => dimsProviderOptions('openai-compatible', 'pplx-embed-v1-0.6b', 2560)).toThrow(AIConfigError);
+  });
+
+  test('init preflight accepts the 4b model at its native 2560 dims (halfvec territory)', () => {
+    const res = resolveSchemaEmbeddingDim({
+      embedding_model: 'perplexity:pplx-embed-v1-4b',
+      embedding_dimensions: 2560,
+    });
+    expect(res).toEqual({
+      ok: true,
+      dim: 2560,
+      model: 'perplexity:pplx-embed-v1-4b',
+      provider: 'perplexity',
+      recipeDefault: 1024,
+    });
+    const bad = resolveSchemaEmbeddingDim({
+      embedding_model: 'perplexity:pplx-embed-v1-4b',
+      embedding_dimensions: 4096,
+    });
+    expect(bad.ok).toBe(false);
+  });
+
+  test('embedding pricing table knows both models', () => {
+    expect(lookupEmbeddingPrice('perplexity:pplx-embed-v1-4b')).toMatchObject({ kind: 'known', pricePerMTok: 0.03 });
+    expect(lookupEmbeddingPrice('perplexity:pplx-embed-v1-0.6b')).toMatchObject({ kind: 'known', pricePerMTok: 0.004 });
+  });
+});
+
+describe('perplexityCompatFetch — int8 wire shim', () => {
+  const realFetch = globalThis.fetch;
+  afterEach(() => {
+    globalThis.fetch = realFetch;
+  });
+
+  test('forces encoding_format=base64_int8 outbound and decodes int8 base64 inbound', async () => {
+    const int8 = new Int8Array([3, -7, 127, -128]);
+    const b64 = Buffer.from(int8.buffer).toString('base64');
+    let sentBody: any;
+    globalThis.fetch = (async (_input: any, init?: RequestInit) => {
+      sentBody = JSON.parse(init!.body as string);
+      return new Response(
+        JSON.stringify({
+          object: 'list',
+          model: 'pplx-embed-v1-4b',
+          data: [{ object: 'embedding', index: 0, embedding: b64 }],
+          usage: { prompt_tokens: 4, total_tokens: 4 },
+        }),
+        { status: 200, headers: { 'content-type': 'application/json' } },
+      );
+    }) as any;
+
+    const resp = await (perplexityCompatFetch as any)('https://api.perplexity.ai/v1/embeddings', {
+      method: 'POST',
+      headers: { 'content-type': 'application/json' },
+      // The AI SDK sends encoding_format:'float' — Perplexity rejects it.
+      body: JSON.stringify({ model: 'pplx-embed-v1-4b', input: ['hi'], encoding_format: 'float', dimensions: 4 }),
+    });
+
+    expect(sentBody.encoding_format).toBe('base64_int8');
+    expect(sentBody.dimensions).toBe(4); // native field, untouched
+    const json = await resp.json();
+    expect(json.data[0].embedding).toEqual([3, -7, 127, -128]);
+    expect(json.usage.prompt_tokens).toBe(4);
+  });
+
+  test('non-JSON and error responses pass through untouched', async () => {
+    globalThis.fetch = (async () =>
+      new Response('nope', { status: 401, headers: { 'content-type': 'text/plain' } })) as any;
+    const resp = await (perplexityCompatFetch as any)('https://api.perplexity.ai/v1/embeddings', {
+      method: 'POST',
+      body: JSON.stringify({ model: 'pplx-embed-v1-4b', input: ['hi'] }),
+    });
+    expect(resp.status).toBe(401);
+    expect(await resp.text()).toBe('nope');
+  });
+});

test/e2e/obsidian-gbrain-safe-index.test.ts

@@ -1,61 +0,0 @@
-/**
- * E2E smoke for skills/obsidian-gbrain-safe-index.
- *
- * Verifies the from-trigger-to-side-effect path that skillify requires:
- * a real user trigger phrase routes to the skill, the resolver/check
- * pipeline treats it as reachable, and the skill file exposes the
- * gbrain commands the workflow actually runs.
- *
- * This stays local-only (no paid embedding, no external API): it asserts
- * the documented safe/import path is present and parseable, not that it
- * mutates a live brain.
- */
-
-import { describe, expect, it } from 'bun:test';
-import { existsSync, readFileSync } from 'fs';
-import { join } from 'path';
-
-const SKILLS = join(import.meta.dir, '..', '..', 'skills');
-const SKILL_MD = join(SKILLS, 'obsidian-gbrain-safe-index', 'SKILL.md');
-const RESOLVER = join(SKILLS, 'RESOLVER.md');
-
-const TRIGGER_PHRASES = [
-  'connect my Obsidian vault to gbrain',
-  'import my vault to gbrain',
-  'sync vault and gbrain',
-  'capture this skill in my vault',
-  'embed gbrain after vault update',
-  'is gbrain synced with my vault',
-];
-
-describe('obsidian-gbrain-safe-index E2E', () => {
-  it('resolver maps real trigger phrasings to the skill', () => {
-    const resolver = readFileSync(RESOLVER, 'utf-8');
-    expect(resolver).toContain('obsidian-gbrain-safe-index/SKILL.md');
-    // Each representative phrase shares a token substring with a resolver row.
-    const rows = resolver
-      .split('\n')
-      .filter((l) => l.includes('obsidian-gbrain-safe-index/SKILL.md'))
-      .join('\n');
-    for (const phrase of TRIGGER_PHRASES) {
-      const hit = phrase
-        .toLowerCase()
-        .split(/\s+/)
-        .some((tok) => tok.length > 3 && rows.toLowerCase().includes(tok));
-      expect(hit, `no resolver token for: ${phrase}`).toBe(true);
-    }
-  });
-
-  it('skill documents the gbrain import-first safe path', () => {
-    const body = readFileSync(SKILL_MD, 'utf-8');
-    expect(body).toContain('gbrain import');
-    expect(body).toContain('--no-embed');
-    expect(body).toContain('gbrain config set search.mode conservative');
-    // Paid gate must be explicit, not a silent default.
-    expect(body).toContain('gbrain embed --stale');
-  });
-
-  it('skill is reachable from the skill tree', () => {
-    expect(existsSync(SKILL_MD)).toBe(true);
-  });
-});

test/e2e/skill-vault-capture-policy.test.ts

@@ -1,52 +0,0 @@
-/**
- * E2E smoke for skills/skill-vault-capture-policy.
- *
- * Verifies the from-trigger-to-side-effect path: a real capture request
- * routes to the skill, and the skill documents the vault navigation
- * obligations (index.md / log.md) that make a capture durable.
- *
- * Local-only: it asserts documented behavior, not live vault writes.
- */
-
-import { describe, expect, it } from 'bun:test';
-import { existsSync, readFileSync } from 'fs';
-import { join } from 'path';
-
-const SKILLS = join(import.meta.dir, '..', '..', 'skills');
-const SKILL_MD = join(SKILLS, 'skill-vault-capture-policy', 'SKILL.md');
-const RESOLVER = join(SKILLS, 'RESOLVER.md');
-
-const TRIGGER_PHRASES = [
-  'save this learning to the vault',
-  'capture this skill in Obsidian',
-  'record this workflow in my notes',
-  'put this setup change in the knowledge base',
-];
-
-describe('skill-vault-capture-policy E2E', () => {
-  it('resolver maps real capture phrasings to the skill', () => {
-    const resolver = readFileSync(RESOLVER, 'utf-8');
-    expect(resolver).toContain('skill-vault-capture-policy/SKILL.md');
-    const rows = resolver
-      .split('\n')
-      .filter((l) => l.includes('skill-vault-capture-policy/SKILL.md'))
-      .join('\n');
-    for (const phrase of TRIGGER_PHRASES) {
-      const hit = phrase
-        .toLowerCase()
-        .split(/\s+/)
-        .some((tok) => tok.length > 3 && rows.toLowerCase().includes(tok));
-      expect(hit, `no resolver token for: ${phrase}`).toBe(true);
-    }
-  });
-
-  it('skill documents index.md and log.md update obligations', () => {
-    const body = readFileSync(SKILL_MD, 'utf-8');
-    expect(body).toContain('index.md');
-    expect(body).toContain('log.md');
-  });
-
-  it('skill is reachable from the skill tree', () => {
-    expect(existsSync(SKILL_MD)).toBe(true);
-  });
-});

test/obsidian-gbrain-safe-index.test.ts

@@ -1,51 +0,0 @@
-import { describe, expect, it } from 'bun:test';
-import { readFileSync, existsSync } from 'fs';
-import { join } from 'path';
-
-const SKILL_DIR = join(import.meta.dir, '..', 'skills', 'obsidian-gbrain-safe-index');
-const SKILL_MD = join(SKILL_DIR, 'SKILL.md');
-const RESOLVER = join(import.meta.dir, '..', 'skills', 'RESOLVER.md');
-
-function parseFrontmatter(raw: string): Record<string, unknown> {
-  const m = raw.match(/^---\n([\s\S]*?)\n---/);
-  if (!m) throw new Error('no frontmatter');
-  const out: Record<string, unknown> = {};
-  for (const line of m[1].split('\n')) {
-    const mm = line.match(/^([a-zA-Z_]+):\s*(.*)$/);
-    if (mm) out[mm[1]] = mm[2].trim();
-  }
-  return out;
-}
-
-describe('obsidian-gbrain-safe-index skill', () => {
-  it('has a SKILL.md with required frontmatter', () => {
-    expect(existsSync(SKILL_MD)).toBe(true);
-    const fm = parseFrontmatter(readFileSync(SKILL_MD, 'utf-8'));
-    expect(fm['name']).toBe('obsidian-gbrain-safe-index');
-    expect(fm['description']).toBeTruthy();
-  });
-
-  it('has the required conformance sections', () => {
-    const body = readFileSync(SKILL_MD, 'utf-8');
-    for (const section of ['## Contract', '## Phases', '## Output Format', '## Anti-Patterns']) {
-      expect(body.includes(section), `missing ${section}`).toBe(true);
-    }
-  });
-
-  it('is registered in RESOLVER.md', () => {
-    expect(existsSync(RESOLVER)).toBe(true);
-    const resolver = readFileSync(RESOLVER, 'utf-8');
-    expect(resolver.includes('obsidian-gbrain-safe-index/SKILL.md')).toBe(true);
-  });
-
-  it('has routing-eval fixtures that exercise real trigger phrasings', () => {
-    const evalPath = join(SKILL_DIR, 'routing-eval.jsonl');
-    expect(existsSync(evalPath)).toBe(true);
-    const lines = readFileSync(evalPath, 'utf-8')
-      .split('\n')
-      .filter((l) => l.trim() && !l.trim().startsWith('//'))
-      .map((l) => JSON.parse(l));
-    const positives = lines.filter((l) => l.expected_skill === 'obsidian-gbrain-safe-index');
-    expect(positives.length).toBeGreaterThanOrEqual(5);
-  });
-});

test/skill-vault-capture-policy.test.ts

@@ -1,64 +0,0 @@
-import { describe, expect, it } from 'bun:test';
-import { readFileSync, existsSync } from 'fs';
-import { join } from 'path';
-import { DEFAULT_PRIVATE_PATTERNS } from '../src/core/skillpack/harvest-lint.ts';
-
-const SKILL_DIR = join(import.meta.dir, '..', 'skills', 'skill-vault-capture-policy');
-const SKILL_MD = join(SKILL_DIR, 'SKILL.md');
-const RESOLVER = join(import.meta.dir, '..', 'skills', 'RESOLVER.md');
-
-function parseFrontmatter(raw: string): Record<string, unknown> {
-  const m = raw.match(/^---\n([\s\S]*?)\n---/);
-  if (!m) throw new Error('no frontmatter');
-  const out: Record<string, unknown> = {};
-  for (const line of m[1].split('\n')) {
-    const mm = line.match(/^([a-zA-Z_]+):\s*(.*)$/);
-    if (mm) out[mm[1]] = mm[2].trim();
-  }
-  return out;
-}
-
-describe('skill-vault-capture-policy skill', () => {
-  it('has a SKILL.md with required frontmatter', () => {
-    expect(existsSync(SKILL_MD)).toBe(true);
-    const fm = parseFrontmatter(readFileSync(SKILL_MD, 'utf-8'));
-    expect(fm['name']).toBe('skill-vault-capture-policy');
-    expect(fm['description']).toBeTruthy();
-  });
-
-  it('has the required conformance sections', () => {
-    const body = readFileSync(SKILL_MD, 'utf-8');
-    for (const section of ['## Contract', '## Phases', '## Output Format', '## Anti-Patterns']) {
-      expect(body.includes(section), `missing ${section}`).toBe(true);
-    }
-  });
-
-  it('is registered in RESOLVER.md', () => {
-    expect(existsSync(RESOLVER)).toBe(true);
-    expect(readFileSync(RESOLVER, 'utf-8').includes('skill-vault-capture-policy/SKILL.md')).toBe(true);
-  });
-
-  it('contains no private user or agent-fork names (privacy rule)', () => {
-    for (const file of [SKILL_MD, join(SKILL_DIR, 'routing-eval.jsonl')]) {
-      const body = readFileSync(file, 'utf-8');
-      // DEFAULT_PRIVATE_PATTERNS[0] is the banned fork-name pattern; sourced
-      // from harvest-lint so this file never contains the literal itself
-      // (scripts/check-privacy.sh would reject it).
-      const forkName = new RegExp(DEFAULT_PRIVATE_PATTERNS[0], 'i');
-      for (const name of [/\bAdam\b/, /\bHermes\b/, /\bHerdr\b/, /\bArk\b/, forkName]) {
-        expect(name.test(body), `private name ${name} in ${file}`).toBe(false);
-      }
-    }
-  });
-
-  it('has routing-eval fixtures', () => {
-    const evalPath = join(SKILL_DIR, 'routing-eval.jsonl');
-    expect(existsSync(evalPath)).toBe(true);
-    const positives = readFileSync(evalPath, 'utf-8')
-      .split('\n')
-      .filter((l) => l.trim() && !l.trim().startsWith('//'))
-      .map((l) => JSON.parse(l))
-      .filter((l) => l.expected_skill === 'skill-vault-capture-policy');
-    expect(positives.length).toBeGreaterThanOrEqual(4);
-  });
-});