Add /jobs page (CI-style UI) + /community v0 stub (#1)

- Add /jobs CI-style page + /community v0 stub (98cf995979cf72e3d8319f72dc0a0ecb58877698)


Co-authored-by: Arthur Zucker <ArthurZ@users.noreply.huggingface.co>

COMMUNITY.md

@@ -0,0 +1,159 @@
+# Slyfox — community roadmap
+
+## Why this matters
+
+Claude Code (and equivalents) generate a uniquely rich signal: a maintainer's
+running internal monologue while they fix bugs, explain weird design choices,
+and triage broken builds. Today, that signal evaporates the moment the user's
+terminal closes. Two things change when maintainers can opt in to publishing
+their traces:
+
+1. **New contributors get a runway.** Someone joining the `transformers` repo
+   for the first time can see how Arthur thinks about a tokenizer fix — without
+   needing a 1:1 onboarding session.
+2. **Open-source models get a training edge.** High-signal expert traces are
+   exactly the kind of data closed labs already farm internally. Putting them
+   in the open shifts the data moat: open-weight models can be fine-tuned on
+   the same kind of corpora that drive frontier-model behavior.
+
+For both of those to work, slyfox needs to be community-centric, not a
+private-dashboard tool. This doc sketches what that looks like.
+
+## North-star user stories
+
+- **Maintainer Maya** runs `hf upload-traces --share community` after a long
+  session on `vllm-project/vllm`. Her bundle goes to a private bucket *and* an
+  opt-in public catalog with a model-fit summary attached.
+- **Contributor Carlos** browses `huggingface/transformers` on the slyfox
+  Space, sees that "Gemma 2 9B" tracks Arthur's emotional cadence the closest
+  on long sessions, picks it for his fork's CI.
+- **Researcher Riya** fine-tunes a 7B model on the public catalog's
+  exported-for-training dataset, posts results back to a leaderboard slot.
+
+## Consent + visibility model
+
+Nothing leaves a maintainer's machine without an explicit flag. The
+defaults are strict; the opt-ins are loud.
+
+| State        | Default | What it means                                                              |
+| ------------ | ------- | -------------------------------------------------------------------------- |
+| `private`    | ✅      | Bucket is theirs only. (Current behavior of `hf upload-traces`.)            |
+| `org`        |         | Bucket is shared within a chosen HF org. Discoverable by org members only.  |
+| `community`  |         | Bundle manifest is published to `HF-slyfox/community-index`; anyone can browse. The trace bucket itself stays per-maintainer-controlled (public / private / read-token). |
+| `training`   |         | Maintainer also tags the bundle with a CC-BY-style flag, signalling that the published traces may be redistributed in training corpora. |
+
+Each escalation is a separate flag (`--share`, `--share-org`, `--publish`,
+`--license cc-by-4.0`) — you never get bumped up a tier silently.
+
+## Phased rollout
+
+### Phase 1 — Catalog (this PR + a follow-up)
+- `/community` page on the Space lists opt-in published runs from
+  `hf://buckets/HF-slyfox/community-index/runs/*.json`.
+- `hf upload-traces` gains `--share community` that writes a minimal
+  manifest entry: `{run_id, owner, repo, bucket, summary, created_at}`.
+- API: `/api/community/runs` — already stubbed in this PR.
+
+### Phase 2 — Browse + remix
+- Each catalog entry has a "Remix" button that resubmits the analysis with the
+  current user's chosen model (calls `/api/analyses` with the source/prefix from
+  the manifest).
+- Per-run page with the EmotionScope orb animation timed to the session's
+  timestamps.
+- Comments / annotations via HF Discussions API on the underlying bucket / dataset.
+
+### Phase 3 — Persona leaderboard
+- For each public bundle, the slyfox Space computes a "persona vector"
+  (average emotion profile, talking-style features, code-language mix) and a
+  per-model fit score across the supported model catalog.
+- `/community/leaderboard?repo=huggingface/transformers` shows which models
+  best match each maintainer's persona on that repo.
+
+### Phase 4 — Training-dataset export
+- Public bundles tagged `--license cc-by-4.0` can be aggregated into a
+  redistributable dataset (`HF-slyfox/training-set`) suitable for SFT/DPO.
+- The exported dataset is *re-scrubbed* (paths, secrets, emails) on top of the
+  uploader-side filter, with a stricter regex set — defense in depth.
+- Periodic releases versioned as `HF-slyfox/training-set:v0.1`, `:v0.2`, etc.
+
+### Phase 5 — Org-level governance
+- Org admins can set `slyfox.community.policy = {allow|deny|approve}` on their
+  HF org; uploads that target a community share check the policy before
+  publishing.
+- Per-repo opt-out (e.g. a private repo never publishes).
+- "Trace of the week" highlighted on the slyfox landing page.
+
+## Data shapes
+
+### Community manifest (one JSON file per published run)
+
+```json
+{
+  "schema_version": 1,
+  "kind": "analysis-run",
+  "run_id": "20260522-184201-a1b2c3",
+  "owner": "ArthurZ",
+  "repo": "huggingface/transformers",
+  "model_id": "google/gemma-2-9b-it",
+  "vectors_ref": "hf://buckets/HF-slyfox/emotion-vectors/google/gemma-2-9b-it/",
+  "source_bucket": "ArthurZ/transformers-traces",
+  "trace_count": 5,
+  "license": "cc-by-4.0",
+  "summary": {
+    "valence_mean": 0.31,
+    "arousal_mean": 0.62,
+    "dominant_emotions": ["focused", "curious", "amused"]
+  },
+  "consent": {
+    "scope": "community",
+    "training_use": true,
+    "issued_at": "2026-05-22T18:42:01Z"
+  },
+  "links": {
+    "results_prefix": "hf://buckets/ArthurZ/transformers-traces/trace-analysis/20260522-184201-a1b2c3/",
+    "discussions": "https://huggingface.co/datasets/ArthurZ/transformers-traces/discussions"
+  }
+}
+```
+
+### Index layout
+
+```
+HF-slyfox/community-index/
+  runs/
+    20260522-184201-a1b2c3.json     # per-run manifests, append-only
+  orgs/
+    huggingface.json                # one-file-per-org summary
+    vllm-project.json
+  leaderboard/
+    huggingface__transformers.json  # cached persona-vs-model scores
+```
+
+## What's already in place vs. what's still on paper
+
+| Piece                              | State          |
+| ---------------------------------- | -------------- |
+| Trace upload with filtering        | ✅ shipped (`hf upload-traces`) |
+| `hf jobs uv` integration           | ✅ this PR     |
+| `/api/community/runs` endpoint     | ✅ stub in this PR |
+| `/community` page                  | ✅ stub in this PR |
+| `--share community` upload flag    | 🚧 next        |
+| Catalog write-side (manifest writer) | 🚧 next      |
+| Remix UI + per-run page            | 🔜 phase 2     |
+| Persona leaderboard                | 🔜 phase 3     |
+| Training-dataset export            | 🔜 phase 4     |
+| Org-level policy / opt-out         | 🔜 phase 5     |
+
+## Open questions
+
+- **Anonymization**: is a `username` ever fine to surface, or do we go fully
+  pseudonymous (`maintainer-1234`) until a user explicitly publishes under
+  their handle? Probably the latter as default with an opt-in toggle.
+- **Quality bar**: should the community catalog filter junk runs (e.g. 1-turn
+  sessions)? Likely yes via a minimum-turns threshold.
+- **Reciprocity**: should publishing be a prerequisite for accessing premium
+  analysis features (GPU time)? Soft incentive vs. paywall — needs a
+  discussion with the org.
+- **Redaction asymmetry**: should the community catalog re-run the filter
+  with a *stricter* regex set than the per-user uploader, on the theory that
+  publishers may have less context to judge what they leak?

Dockerfile

@@ -9,7 +9,7 @@ WORKDIR /app
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
-COPY app.py index.html ./
+COPY app.py index.html jobs.html community.html ./
 
 # HF Spaces require listening on $PORT (default 7860).
 ENV PORT=7860

README.md

@@ -67,3 +67,49 @@ HF_TOKEN=hf_… uvicorn app:app --reload --port 7860
 ```
 
 Then visit `http://localhost:7860/?api=http://localhost:8000/v1`.
+
+## Jobs page
+
+A second route, `/jobs`, exposes two operations that submit Hugging Face Jobs
+on behalf of the user. Both run via `HfApi.run_uv_job` and report CI-style
+status back into the UI.
+
+| Operation | Submits | Output |
+| --- | --- | --- |
+| **Request a new model** | `jobs_scripts/extract_vectors.py` | `hf://buckets/HF-slyfox/emotion-vectors/<model_id>/` |
+| **Analyze traces** | `jobs_scripts/analyze_traces.py` | `hf://buckets/<source>/trace-analysis/<run_id>/` |
+
+API surface (additive — none of the existing `/api/chats/*` routes change):
+
+```
+POST /api/models/extract            { model_id, flavor? }
+POST /api/analyses                  { source, prefix?, n_traces, model_id, flavor? }
+GET  /api/jobs                      list slyfox-submitted jobs with live status
+GET  /api/jobs/{job_id}             one job's record + live status
+GET  /api/jobs/{job_id}/logs        tail of the job's stdout/stderr
+POST /api/jobs/{job_id}/cancel      cancel a running/queued job
+```
+
+Bucket layout extension:
+
+```
+SLYFOX_BUCKET/
+  ...existing chat layout...
+  jobs/<job_id>.json     # sidecar record for each submitted job
+```
+
+Extra env vars:
+
+| Variable | Default | Description |
+| --- | --- | --- |
+| `SLYFOX_JOBS_REF`         | `main` | Git ref of `huggingface/slyfox` to pull UV scripts from. |
+| `SLYFOX_VECTORS_BUCKET`   | `HF-slyfox/emotion-vectors` | Where extracted vectors land. |
+| `SLYFOX_COMMUNITY_BUCKET` | `HF-slyfox/community-index` | Public index of opt-in shared runs. |
+
+## Community (v0)
+
+A third route, `/community`, is a placeholder for the upcoming public catalog
+of opt-in shared analysis runs. The `/api/community/runs` endpoint already
+reads from `SLYFOX_COMMUNITY_BUCKET`; the publish-side flow (`hf upload-traces
+--share community`) is on the roadmap. See [`COMMUNITY.md`](./COMMUNITY.md)
+for the full design.

app.py

@@ -4,6 +4,7 @@ Layout in the bucket (one file per chat):
 
     users/<user_id>/chats/<chat_id>.json
     users/<user_id>/index.json          # ordered list of chat ids + titles + topics
+    jobs/<job_id>.json                  # slyfox-submitted hf jobs (extract / analyze)
 
 `<user_id>` is currently an anonymous UUID minted in the browser (stored in
 localStorage) and sent as `X-User-Id`. This is intentionally minimal — swap for
@@ -17,6 +18,7 @@ import json
 import logging
 import os
 import re
+from datetime import datetime, timezone
 from pathlib import Path
 from typing import Any
 
@@ -33,11 +35,20 @@ log = logging.getLogger("slyfox")
 BUCKET_ID = os.getenv("SLYFOX_BUCKET", "HF-slyfox/slyfox-chats")
 HF_TOKEN = os.getenv("HF_TOKEN")  # Spaces inject this automatically.
 
+# Job submission config — URLs of UV scripts that run on hf jobs infra.
+SLYFOX_JOBS_REF = os.getenv("SLYFOX_JOBS_REF", "arthur-work")  # branch / tag of the slyfox repo (`main` once promoted)
+JOBS_SCRIPT_BASE = f"https://raw.githubusercontent.com/huggingface/slyfox/{SLYFOX_JOBS_REF}/jobs_scripts"
+EXTRACT_SCRIPT_URL = f"{JOBS_SCRIPT_BASE}/extract_vectors.py"
+ANALYZE_SCRIPT_URL = f"{JOBS_SCRIPT_BASE}/analyze_traces.py"
+VECTORS_BUCKET = os.getenv("SLYFOX_VECTORS_BUCKET", "HF-slyfox/emotion-vectors")
+
 api = HfApi(token=HF_TOKEN)
 
 ROOT = Path(__file__).parent
 USER_RE = re.compile(r"^[A-Za-z0-9_-]{8,64}$")
 CHAT_RE = re.compile(r"^[A-Za-z0-9_-]{8,64}$")
+MODEL_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._\-]*/[A-Za-z0-9][A-Za-z0-9._\-]*$")
+BUCKET_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._\-]*/[A-Za-z0-9][A-Za-z0-9._\-]*$")
 
 
 def _user_prefix(user_id: str) -> str:
@@ -66,8 +77,6 @@ def _read_json(remote_path: str) -> Any | None:
         log.warning("metadata failed for %s: %s", remote_path, e)
         return None
 
-    # `download_bucket_files` writes to disk; pull bytes via the file handle utility.
-    # Easiest cross-version approach: download to a temp path and read it back.
     import tempfile
 
     with tempfile.TemporaryDirectory() as tmp:
@@ -87,6 +96,24 @@ def _write_json(remote_path: str, payload: Any) -> None:
     api.batch_bucket_files(BUCKET_ID, add=[(data, remote_path)])
 
 
+def _list_json(prefix: str) -> list[dict]:
+    """List JSON objects directly under `prefix/` in the bucket."""
+    try:
+        files = list(api.list_bucket_files(BUCKET_ID, path=prefix))
+    except Exception as e:  # noqa: BLE001
+        log.warning("list failed for %s: %s", prefix, e)
+        return []
+    out: list[dict] = []
+    for f in files:
+        path = getattr(f, "path", None) or getattr(f, "rfilename", None) or str(f)
+        if not path.endswith(".json"):
+            continue
+        data = _read_json(path)
+        if isinstance(data, dict):
+            out.append(data)
+    return out
+
+
 def _delete(remote_path: str) -> None:
     try:
         api.batch_bucket_files(BUCKET_ID, delete=[remote_path])
@@ -94,9 +121,18 @@ def _delete(remote_path: str) -> None:
         log.warning("delete failed for %s: %s", remote_path, e)
 
 
+def _now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat(timespec="seconds")
+
+
 app = FastAPI(title="Slyfox")
 
 
+# =============================================================================
+# Existing chat persistence API (unchanged)
+# =============================================================================
+
+
 @app.get("/api/chats")
 def list_chats(x_user_id: str = Header(...)) -> JSONResponse:
     """Return the per-user index: [{id, title, topic, updated_at}, ...]."""
@@ -115,13 +151,11 @@ def get_chat(chat_id: str, x_user_id: str = Header(...)) -> JSONResponse:
 @app.put("/api/chats/{chat_id}")
 async def put_chat(chat_id: str, request: Request, x_user_id: str = Header(...)) -> JSONResponse:
     body = await request.json()
-    # Expected shape: {id, title, topic, history: [...], updated_at}
     if body.get("id") != chat_id:
         raise HTTPException(400, "chat id mismatch")
 
     _write_json(_chat_path(x_user_id, chat_id), body)
 
-    # Update the per-user index (one row per chat, newest first).
     idx = _read_json(_index_path(x_user_id)) or []
     idx = [e for e in idx if e.get("id") != chat_id]
     idx.insert(
@@ -151,7 +185,267 @@ def healthz() -> dict:
     return {"ok": True, "bucket": BUCKET_ID}
 
 
+# =============================================================================
+# Jobs API — emotion-vector extraction + trace analysis via `hf jobs`
+# =============================================================================
+#
+# Two job kinds, both submitted via `HfApi.run_uv_job`:
+#
+#   extract : take a model_id, run EmotionScope's vector extraction, write
+#             the 20 emotion direction vectors to a bucket under
+#             <VECTORS_BUCKET>/<model_id>/
+#
+#   analyze : take a traces source (bucket+prefix) and a model_id, replay
+#             sessions through the model, project each turn's residual stream
+#             onto the emotion vectors, write per-trace parquet results under
+#             <source>/trace-analysis/<run_id>/
+#
+# A small sidecar record is written to the chats bucket under `jobs/<job_id>.json`
+# so the Space can list its own jobs. Live status / logs are pulled directly
+# from the HF jobs API on demand.
+
+
+def _job_record(job_id: str, *, kind: str, params: dict, requester: str) -> dict:
+    return {
+        "job_id": job_id,
+        "kind": kind,
+        "params": params,
+        "requester": requester,
+        "submitted_at": _now_iso(),
+    }
+
+
+def _hf_job_status(job_id: str) -> dict:
+    """Live status snapshot for one job. Falls back gracefully on errors."""
+    try:
+        info = api.inspect_job(job_id)
+    except Exception as e:  # noqa: BLE001
+        return {"job_id": job_id, "status": "unknown", "error": str(e)}
+    started = getattr(info, "started_at", None)
+    ended = getattr(info, "ended_at", None)
+    elapsed = None
+    if started:
+        end = ended or datetime.now(timezone.utc)
+        try:
+            elapsed = int((end - started).total_seconds())
+        except Exception:  # noqa: BLE001
+            elapsed = None
+    return {
+        "job_id": job_id,
+        "status": getattr(info, "status", None) or getattr(info, "stage", "unknown"),
+        "flavor": getattr(info, "flavor", None),
+        "started_at": started.isoformat() if hasattr(started, "isoformat") else started,
+        "ended_at": ended.isoformat() if hasattr(ended, "isoformat") else ended,
+        "elapsed_s": elapsed,
+    }
+
+
+@app.post("/api/models/extract")
+async def request_model_extract(request: Request, x_user_id: str = Header(...)) -> JSONResponse:
+    """Submit an emotion-vector extraction job for a HuggingFace model."""
+    _user_prefix(x_user_id)
+    body = await request.json()
+    model_id = (body.get("model_id") or "").strip()
+    flavor = body.get("flavor") or "a10g-large"
+    if not MODEL_RE.match(model_id):
+        raise HTTPException(400, "model_id must look like 'org/name'")
+
+    output = f"hf://buckets/{VECTORS_BUCKET}/{model_id}/"
+    job = api.run_uv_job(
+        script=EXTRACT_SCRIPT_URL,
+        script_args=["--model", model_id, "--output", output],
+        dependencies=[
+            "transformers>=4.45",
+            "torch",
+            "huggingface_hub>=1.7",
+            "scikit-learn",
+            "safetensors",
+            "numpy",
+        ],
+        flavor=flavor,
+        timeout="45m",
+        labels={
+            "slyfox.io/type": "extract",
+            "slyfox.io/model": model_id,
+            "slyfox.io/requester": x_user_id,
+        },
+    )
+    record = _job_record(
+        job.id,
+        kind="extract",
+        params={"model_id": model_id, "flavor": flavor, "output": output},
+        requester=x_user_id,
+    )
+    _write_json(f"jobs/{job.id}.json", record)
+    return JSONResponse(record)
+
+
+@app.post("/api/analyses")
+async def request_analysis(request: Request, x_user_id: str = Header(...)) -> JSONResponse:
+    """Submit a trace-analysis job: replay sessions through a model and emit emotion trajectories."""
+    _user_prefix(x_user_id)
+    body = await request.json()
+    source = (body.get("source") or "").strip()  # bucket id, e.g. "HF-slyfox/traces"
+    prefix = (body.get("prefix") or "").strip().strip("/")
+    n_traces = int(body.get("n_traces") or 5)
+    model_id = (body.get("model_id") or "").strip()
+    flavor = body.get("flavor") or "a10g-large"
+
+    if not BUCKET_RE.match(source):
+        raise HTTPException(400, "source must look like 'namespace/bucket'")
+    if not MODEL_RE.match(model_id):
+        raise HTTPException(400, "model_id must look like 'org/name'")
+    if n_traces < 1 or n_traces > 100:
+        raise HTTPException(400, "n_traces must be 1..100")
+
+    source_url = f"hf://buckets/{source}/{prefix}/" if prefix else f"hf://buckets/{source}/"
+    output_url = f"hf://buckets/{source}/trace-analysis/"
+    vectors_url = f"hf://buckets/{VECTORS_BUCKET}/{model_id}/"
+    job = api.run_uv_job(
+        script=ANALYZE_SCRIPT_URL,
+        script_args=[
+            "--source", source_url,
+            "--model", model_id,
+            "--vectors", vectors_url,
+            "--n-traces", str(n_traces),
+            "--output", output_url,
+        ],
+        dependencies=[
+            "transformers>=4.45",
+            "torch",
+            "huggingface_hub>=1.7",
+            "safetensors",
+            "numpy",
+            "pyarrow",
+        ],
+        flavor=flavor,
+        timeout="2h",
+        labels={
+            "slyfox.io/type": "analyze",
+            "slyfox.io/model": model_id,
+            "slyfox.io/source": source,
+            "slyfox.io/requester": x_user_id,
+        },
+    )
+    record = _job_record(
+        job.id,
+        kind="analyze",
+        params={
+            "source": source,
+            "prefix": prefix,
+            "n_traces": n_traces,
+            "model_id": model_id,
+            "flavor": flavor,
+            "output": output_url,
+            "vectors": vectors_url,
+        },
+        requester=x_user_id,
+    )
+    _write_json(f"jobs/{job.id}.json", record)
+    return JSONResponse(record)
+
+
+@app.get("/api/jobs")
+def list_slyfox_jobs() -> JSONResponse:
+    """List slyfox-submitted jobs (newest first) with live status merged in."""
+    records = _list_json("jobs")
+    records.sort(key=lambda r: r.get("submitted_at", ""), reverse=True)
+    out: list[dict] = []
+    for r in records:
+        out.append({**r, "live": _hf_job_status(r["job_id"])})
+    return JSONResponse(out)
+
+
+@app.get("/api/jobs/{job_id}")
+def get_job(job_id: str) -> JSONResponse:
+    record = _read_json(f"jobs/{job_id}.json") or {"job_id": job_id}
+    return JSONResponse({**record, "live": _hf_job_status(job_id)})
+
+
+@app.get("/api/jobs/{job_id}/logs")
+def get_job_logs(job_id: str, tail: int = 200) -> JSONResponse:
+    try:
+        chunks = list(api.fetch_job_logs(job_id))
+    except Exception as e:  # noqa: BLE001
+        raise HTTPException(404, f"logs unavailable: {e}")
+    text = "".join(chunks)
+    lines = text.splitlines()
+    if tail and tail > 0:
+        lines = lines[-tail:]
+    return JSONResponse({"job_id": job_id, "lines": lines})
+
+
+@app.post("/api/jobs/{job_id}/cancel")
+def cancel_slyfox_job(job_id: str, x_user_id: str = Header(...)) -> JSONResponse:
+    _user_prefix(x_user_id)
+    try:
+        api.cancel_job(job_id)
+    except Exception as e:  # noqa: BLE001
+        raise HTTPException(404, f"cancel failed: {e}")
+    return JSONResponse({"ok": True, "job_id": job_id})
+
+
+@app.get("/jobs", include_in_schema=False)
+def jobs_page() -> FileResponse:
+    return FileResponse(ROOT / "jobs.html")
+
+
+# =============================================================================
+# Community API — placeholder surface for org-level / public trace sharing.
+# =============================================================================
+#
+# v0 stubs reserve the URL shape so future PRs can land without renaming.
+# The intent (see COMMUNITY.md) is that orgs and users *opt in* to publishing
+# their trace bundles + analysis runs to a public index, which the Space
+# surfaces here so anyone can browse other maintainers' personas, remix
+# analyses with their own model picks, and (eventually) export
+# fine-tuning-friendly datasets from the published traces.
+
+COMMUNITY_INDEX_BUCKET = os.getenv("SLYFOX_COMMUNITY_BUCKET", "HF-slyfox/community-index")
+
+
+@app.get("/api/community/runs")
+def community_runs() -> JSONResponse:
+    """Return the public list of opt-in published analysis runs.
+
+    v0: returns whatever has been written to
+        hf://buckets/<SLYFOX_COMMUNITY_BUCKET>/runs/*.json
+    Each entry is a manifest with {run_id, model_id, source, owner, summary}.
+    """
+    try:
+        files = list(api.list_bucket_files(COMMUNITY_INDEX_BUCKET, path="runs"))
+    except Exception as e:  # noqa: BLE001
+        log.info("community index not available yet: %s", e)
+        return JSONResponse({"runs": [], "note": "community index not initialized"})
+    runs: list[dict] = []
+    for f in files:
+        path = getattr(f, "path", None) or getattr(f, "rfilename", None) or str(f)
+        if not path.endswith(".json"):
+            continue
+        # Read each manifest. The community bucket is *public*, so anonymous
+        # reads should work; we still go via HfApi for consistency.
+        import tempfile
+
+        with tempfile.TemporaryDirectory() as tmp:
+            local = Path(tmp) / "m.json"
+            try:
+                api.download_bucket_files(COMMUNITY_INDEX_BUCKET, files=[(path, str(local))])
+                runs.append(json.loads(local.read_text()))
+            except Exception as e:  # noqa: BLE001
+                log.warning("skip community entry %s: %s", path, e)
+    return JSONResponse({"runs": runs})
+
+
+@app.get("/community", include_in_schema=False)
+def community_page() -> FileResponse:
+    return FileResponse(ROOT / "community.html")
+
+
+# =============================================================================
 # Static frontend — mounted last so /api/* routes win.
+# =============================================================================
+
+
 @app.get("/")
 def index() -> FileResponse:
     return FileResponse(ROOT / "index.html")

community.html

@@ -0,0 +1,120 @@
+<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8" />
+<meta name="viewport" content="width=device-width,initial-scale=1" />
+<title>Slyfox · Community</title>
+<style>
+  :root {
+    --bg: #0d1117; --panel: #161b22; --panel-2: #1c2128; --border: #30363d;
+    --text: #e6edf3; --muted: #8b949e; --accent: #2f81f7;
+    --ok: #3fb950; --warn: #d29922; --err: #f85149;
+    --mono: ui-monospace, SFMono-Regular, "SF Mono", Menlo, Consolas, monospace;
+  }
+  * { box-sizing: border-box; }
+  html, body { background: var(--bg); color: var(--text); margin: 0; font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif; }
+  a { color: var(--accent); text-decoration: none; }
+  a:hover { text-decoration: underline; }
+  header { display: flex; align-items: center; justify-content: space-between; padding: 14px 24px; border-bottom: 1px solid var(--border); }
+  header h1 { font-size: 16px; margin: 0; font-weight: 600; }
+  header .nav a { color: var(--muted); margin-left: 18px; font-size: 14px; }
+  header .nav a.active { color: var(--text); }
+  main { max-width: 1100px; margin: 0 auto; padding: 24px; }
+  .panel { background: var(--panel); border: 1px solid var(--border); border-radius: 8px; padding: 18px 20px; margin-bottom: 18px; }
+  .panel h2 { margin: 0 0 10px; font-size: 14px; font-weight: 600; }
+  .panel p { color: var(--muted); font-size: 13px; line-height: 1.6; margin: 0 0 10px; }
+  .badge { display: inline-block; padding: 2px 8px; border-radius: 999px; font-size: 11px; background: var(--panel-2); color: var(--muted); border: 1px solid var(--border); margin-right: 6px; }
+  .runs { display: grid; gap: 10px; }
+  .run { background: var(--panel-2); border: 1px solid var(--border); border-radius: 8px; padding: 12px 14px; }
+  .run h3 { margin: 0; font-size: 13px; font-weight: 500; }
+  .run h3 .id { color: var(--muted); font-family: var(--mono); font-size: 11px; margin-left: 6px; }
+  .run .meta { color: var(--muted); font-size: 11px; margin-top: 4px; }
+  .empty { padding: 24px; text-align: center; color: var(--muted); font-size: 13px; }
+  pre { background: #0a0d12; border: 1px solid var(--border); padding: 12px; border-radius: 6px; font-family: var(--mono); font-size: 12px; overflow: auto; color: #c9d1d9; }
+</style>
+</head>
+<body>
+<header>
+  <h1>🦊 Slyfox · Community</h1>
+  <nav class="nav">
+    <a href="/">Chat</a>
+    <a href="/jobs">Jobs</a>
+    <a href="/community" class="active">Community</a>
+  </nav>
+</header>
+
+<main>
+  <section class="panel">
+    <h2>What this is <span class="badge">v0 stub</span></h2>
+    <p>
+      Slyfox is set up so any maintainer can publish their Claude Code traces +
+      analysis runs to a public catalog. The idea: shared traces are a high-signal
+      training and tuning corpus for open-source models, and they help
+      contributors get up to speed on a repo without needing direct access to
+      the original maintainer.
+    </p>
+    <p>
+      This page lists opt-in <b>public analysis runs</b> — manifests that
+      maintainers chose to share. The browse / remix / leaderboard UX is being
+      built; the wiring below is already in place.
+    </p>
+    <p>
+      See <code>COMMUNITY.md</code> in the slyfox repo for the full roadmap and
+      the consent / visibility model.
+    </p>
+  </section>
+
+  <section class="panel">
+    <h2>Public runs</h2>
+    <div id="runs" class="runs"><div class="empty">loading…</div></div>
+  </section>
+
+  <section class="panel">
+    <h2>How to publish</h2>
+    <p>From a clone of a repo whose maintainer has set up a slyfox bucket:</p>
+<pre>hf upload-traces \
+  --bucket &lt;org&gt;/&lt;repo&gt;-traces \
+  --share community           # opt-in: writes a manifest to HF-slyfox/community-index
+
+# Then from this Space, run an analysis with `--publish` to add the run to the
+# public catalog. Anyone can browse, fork, or download.</pre>
+    <p style="color: var(--warn); font-size: 12px;">
+      Note: <code>--share</code> / <code>--publish</code> are part of the
+      roadmap, not yet implemented. Don't paste secrets you don't want indexed.
+    </p>
+  </section>
+</main>
+
+<script>
+"use strict";
+async function load() {
+  const root = document.getElementById("runs");
+  try {
+    const res = await fetch("/api/community/runs");
+    const data = await res.json();
+    const runs = data.runs || [];
+    if (!runs.length) {
+      root.innerHTML = `<div class="empty">no public runs yet — be the first to publish.</div>`;
+      return;
+    }
+    root.innerHTML = runs.map(r => `
+      <div class="run">
+        <h3>${esc(r.title || r.model_id || r.run_id)}<span class="id">${esc(r.run_id || "")}</span></h3>
+        <div class="meta">
+          ${r.owner ? `<span class="badge">${esc(r.owner)}</span>` : ""}
+          ${r.model_id ? `<span class="badge">${esc(r.model_id)}</span>` : ""}
+          ${r.source ? `<span class="badge">${esc(r.source)}</span>` : ""}
+          ${r.completed_at || ""}
+        </div>
+      </div>
+    `).join("");
+  } catch (e) {
+    root.innerHTML = `<div class="empty">failed to load: ${esc(e.message)}</div>`;
+  }
+}
+function esc(s) { return String(s).replace(/[&<>"']/g, c => ({"&":"&amp;","<":"&lt;",">":"&gt;",'"':"&quot;","'":"&#39;"}[c])); }
+load();
+setInterval(load, 30000);
+</script>
+</body>
+</html>

jobs.html

@@ -0,0 +1,428 @@
+<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8" />
+<meta name="viewport" content="width=device-width,initial-scale=1" />
+<title>Slyfox · Jobs</title>
+<style>
+  :root {
+    --bg: #0d1117;
+    --panel: #161b22;
+    --panel-2: #1c2128;
+    --border: #30363d;
+    --text: #e6edf3;
+    --muted: #8b949e;
+    --accent: #2f81f7;
+    --ok: #3fb950;
+    --warn: #d29922;
+    --err: #f85149;
+    --pending: #6e7681;
+    --running: #58a6ff;
+    --mono: ui-monospace, SFMono-Regular, "SF Mono", Menlo, Consolas, monospace;
+  }
+  * { box-sizing: border-box; }
+  html, body { background: var(--bg); color: var(--text); margin: 0; font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif; }
+  a { color: var(--accent); text-decoration: none; }
+  a:hover { text-decoration: underline; }
+
+  header {
+    display: flex; align-items: center; justify-content: space-between;
+    padding: 14px 24px; border-bottom: 1px solid var(--border);
+  }
+  header h1 { font-size: 16px; margin: 0; font-weight: 600; letter-spacing: 0.2px; }
+  header .nav a { color: var(--muted); margin-left: 18px; font-size: 14px; }
+  header .nav a.active { color: var(--text); }
+
+  main { max-width: 1100px; margin: 0 auto; padding: 24px; }
+
+  .panel {
+    background: var(--panel); border: 1px solid var(--border); border-radius: 8px;
+    padding: 18px 20px; margin-bottom: 18px;
+  }
+  .panel h2 {
+    margin: 0 0 12px; font-size: 14px; font-weight: 600; color: var(--text);
+    display: flex; align-items: center; gap: 8px;
+  }
+  .panel h2 .hint { color: var(--muted); font-weight: 400; font-size: 12px; }
+
+  .grid { display: grid; gap: 12px; grid-template-columns: repeat(auto-fit, minmax(220px, 1fr)); }
+  label { display: block; font-size: 12px; color: var(--muted); margin-bottom: 4px; }
+  input[type=text], input[type=number], select {
+    width: 100%;
+    background: var(--panel-2); border: 1px solid var(--border); border-radius: 6px;
+    color: var(--text); padding: 8px 10px; font-size: 13px; font-family: inherit;
+  }
+  input:focus, select:focus { outline: none; border-color: var(--accent); }
+  button {
+    background: var(--accent); color: white; border: 0; border-radius: 6px;
+    padding: 8px 14px; font-size: 13px; cursor: pointer; font-weight: 500;
+  }
+  button:hover { filter: brightness(1.1); }
+  button.secondary { background: var(--panel-2); color: var(--text); border: 1px solid var(--border); }
+  button.danger { background: var(--err); }
+  button[disabled] { opacity: 0.5; cursor: not-allowed; }
+
+  .row-actions { display: flex; gap: 10px; margin-top: 14px; align-items: center; }
+  .row-actions .filler { flex: 1; }
+  .toast { font-size: 12px; color: var(--muted); }
+  .toast.ok { color: var(--ok); }
+  .toast.err { color: var(--err); }
+
+  /* Job list — CI-style rows */
+  .jobs { border-radius: 8px; border: 1px solid var(--border); overflow: hidden; }
+  .job-row {
+    display: grid; grid-template-columns: 28px 24px 1fr auto auto;
+    gap: 12px; align-items: center;
+    padding: 12px 14px; background: var(--panel); border-bottom: 1px solid var(--border);
+    cursor: pointer; transition: background 0.1s;
+  }
+  .job-row:last-child { border-bottom: 0; }
+  .job-row:hover { background: var(--panel-2); }
+  .job-row.expanded { background: var(--panel-2); }
+
+  .dot {
+    width: 12px; height: 12px; border-radius: 50%;
+    display: inline-block; flex-shrink: 0;
+    box-shadow: 0 0 0 2px var(--panel);
+  }
+  .dot.queued, .dot.pending { background: var(--pending); }
+  .dot.running { background: var(--running); animation: pulse 1.4s ease-in-out infinite; }
+  .dot.completed, .dot.success { background: var(--ok); }
+  .dot.failed, .dot.error { background: var(--err); }
+  .dot.cancelled, .dot.canceled { background: var(--warn); }
+  .dot.unknown { background: var(--pending); opacity: 0.5; }
+  @keyframes pulse { 0%, 100% { opacity: 1; } 50% { opacity: 0.4; } }
+
+  .kind { font-size: 16px; opacity: 0.9; text-align: center; }
+  .job-title { font-size: 13px; color: var(--text); font-weight: 500; }
+  .job-title .id { color: var(--muted); font-family: var(--mono); margin-left: 6px; font-size: 11px; }
+  .job-sub { color: var(--muted); font-size: 11px; margin-top: 2px; }
+  .job-meta { color: var(--muted); font-size: 12px; font-family: var(--mono); text-align: right; }
+  .job-meta .status { display: block; color: var(--text); font-size: 11px; text-transform: uppercase; letter-spacing: 0.4px; }
+  .chevron { color: var(--muted); font-size: 14px; transition: transform 0.15s; }
+  .job-row.expanded .chevron { transform: rotate(90deg); }
+
+  .job-detail {
+    background: #0a0d12; border-bottom: 1px solid var(--border);
+    padding: 14px 18px; font-family: var(--mono); font-size: 12px;
+    display: none;
+  }
+  .job-row.expanded + .job-detail { display: block; }
+
+  .job-detail .params { color: var(--muted); margin-bottom: 10px; line-height: 1.6; }
+  .job-detail .params b { color: var(--text); font-weight: 500; }
+  .job-detail .logs {
+    background: #000; border: 1px solid var(--border); border-radius: 6px;
+    padding: 10px 12px; max-height: 280px; overflow: auto; white-space: pre-wrap;
+    color: #c9d1d9; line-height: 1.5;
+  }
+  .job-detail .actions { display: flex; gap: 8px; margin-top: 10px; }
+
+  .empty { padding: 24px; text-align: center; color: var(--muted); font-size: 13px; }
+</style>
+</head>
+<body>
+<header>
+  <h1>🦊 Slyfox · Jobs</h1>
+  <nav class="nav">
+    <a href="/">Chat</a>
+    <a href="/jobs" class="active">Jobs</a>
+    <a href="/community">Community</a>
+  </nav>
+</header>
+
+<main>
+  <!-- Request model -->
+  <section class="panel" id="panel-extract">
+    <h2>Request a new model <span class="hint">runs emotion-vector extraction on hf jobs</span></h2>
+    <div class="grid">
+      <div>
+        <label for="ex-model">model id</label>
+        <input id="ex-model" type="text" placeholder="Qwen/Qwen2.5-7B-Instruct" />
+      </div>
+      <div>
+        <label for="ex-flavor">hardware</label>
+        <select id="ex-flavor">
+          <option value="a10g-large">a10g-large (default)</option>
+          <option value="a10g-small">a10g-small</option>
+          <option value="a100-large">a100-large</option>
+          <option value="cpu-upgrade">cpu-upgrade</option>
+        </select>
+      </div>
+    </div>
+    <div class="row-actions">
+      <button id="ex-submit">Submit extraction</button>
+      <span class="toast" id="ex-toast"></span>
+    </div>
+  </section>
+
+  <!-- Analyze traces -->
+  <section class="panel" id="panel-analyze">
+    <h2>Analyze traces <span class="hint">replays sessions through a model, emits per-turn emotion trajectories</span></h2>
+    <div class="grid">
+      <div>
+        <label for="an-source">traces bucket</label>
+        <input id="an-source" type="text" placeholder="HF-slyfox/traces" value="HF-slyfox/traces" />
+      </div>
+      <div>
+        <label for="an-prefix">prefix (optional)</label>
+        <input id="an-prefix" type="text" placeholder="Arthurs-beast/-Users-...-transformers" />
+      </div>
+      <div>
+        <label for="an-n">num traces</label>
+        <input id="an-n" type="number" min="1" max="100" value="5" />
+      </div>
+      <div>
+        <label for="an-model">model</label>
+        <input id="an-model" type="text" placeholder="google/gemma-2-2b-it" />
+      </div>
+      <div>
+        <label for="an-flavor">hardware</label>
+        <select id="an-flavor">
+          <option value="a10g-large">a10g-large</option>
+          <option value="a10g-small">a10g-small</option>
+          <option value="a100-large">a100-large</option>
+        </select>
+      </div>
+    </div>
+    <div class="row-actions">
+      <button id="an-submit">Submit analysis</button>
+      <span class="toast" id="an-toast"></span>
+    </div>
+  </section>
+
+  <!-- Job list -->
+  <section>
+    <h2 style="font-size: 14px; font-weight: 600; margin: 0 0 10px;">Recent runs</h2>
+    <div class="jobs" id="jobs"></div>
+  </section>
+</main>
+
+<script>
+"use strict";
+
+// --------------------------------------------------------------------------
+// Tiny per-browser user id, persisted in localStorage. Mirrors index.html.
+// --------------------------------------------------------------------------
+
+const USER_KEY = "slyfox.user_id";
+function userId() {
+  let id = localStorage.getItem(USER_KEY);
+  if (!id) {
+    id = "u-" + crypto.randomUUID().replaceAll("-", "").slice(0, 16);
+    localStorage.setItem(USER_KEY, id);
+  }
+  return id;
+}
+
+async function api(path, opts = {}) {
+  const headers = { "X-User-Id": userId(), ...(opts.headers || {}) };
+  if (opts.body && typeof opts.body !== "string") {
+    headers["Content-Type"] = "application/json";
+    opts.body = JSON.stringify(opts.body);
+  }
+  const res = await fetch(path, { ...opts, headers });
+  if (!res.ok) {
+    let err;
+    try { err = (await res.json()).detail || res.statusText; } catch { err = res.statusText; }
+    throw new Error(err);
+  }
+  return res.json();
+}
+
+// --------------------------------------------------------------------------
+// Submit handlers
+// --------------------------------------------------------------------------
+
+function bindSubmit(buttonId, toastId, build, endpoint) {
+  const btn = document.getElementById(buttonId);
+  const toast = document.getElementById(toastId);
+  btn.addEventListener("click", async () => {
+    btn.disabled = true;
+    toast.textContent = "submitting…";
+    toast.className = "toast";
+    try {
+      const body = build();
+      const job = await api(endpoint, { method: "POST", body });
+      toast.textContent = "queued: " + job.job_id;
+      toast.className = "toast ok";
+      refreshJobs();
+    } catch (e) {
+      toast.textContent = "✗ " + e.message;
+      toast.className = "toast err";
+    } finally {
+      btn.disabled = false;
+    }
+  });
+}
+
+bindSubmit("ex-submit", "ex-toast", () => ({
+  model_id: document.getElementById("ex-model").value.trim(),
+  flavor: document.getElementById("ex-flavor").value,
+}), "/api/models/extract");
+
+bindSubmit("an-submit", "an-toast", () => ({
+  source: document.getElementById("an-source").value.trim(),
+  prefix: document.getElementById("an-prefix").value.trim(),
+  n_traces: parseInt(document.getElementById("an-n").value, 10),
+  model_id: document.getElementById("an-model").value.trim(),
+  flavor: document.getElementById("an-flavor").value,
+}), "/api/analyses");
+
+// --------------------------------------------------------------------------
+// Job list rendering
+// --------------------------------------------------------------------------
+
+const jobsRoot = document.getElementById("jobs");
+const expanded = new Set();
+const logCache = new Map();
+let logTimer = null;
+
+const KIND_ICON = { extract: "🧠", analyze: "🧪" };
+
+function statusClass(s) {
+  if (!s) return "unknown";
+  s = String(s).toLowerCase();
+  if (s.includes("run")) return "running";
+  if (s.includes("queue") || s.includes("pend") || s.includes("start")) return "queued";
+  if (s.includes("complete") || s.includes("succ") || s.includes("done")) return "completed";
+  if (s.includes("fail") || s.includes("err")) return "failed";
+  if (s.includes("cancel")) return "cancelled";
+  return "unknown";
+}
+
+function fmtElapsed(s) {
+  if (!s && s !== 0) return "—";
+  const m = Math.floor(s / 60), sec = s % 60;
+  if (m === 0) return sec + "s";
+  if (m < 60) return m + "m " + sec + "s";
+  const h = Math.floor(m / 60);
+  return h + "h " + (m % 60) + "m";
+}
+
+function jobTitle(j) {
+  if (j.kind === "extract") return "Extract " + (j.params?.model_id || "?");
+  if (j.kind === "analyze") return "Analyze " + (j.params?.source || "?") + " · " + (j.params?.model_id || "?");
+  return j.job_id;
+}
+
+function jobSubtitle(j) {
+  if (j.kind === "extract") return "→ " + (j.params?.output || "");
+  if (j.kind === "analyze") {
+    const n = j.params?.n_traces ?? "?";
+    return n + " traces → " + (j.params?.output || "");
+  }
+  return "";
+}
+
+function paramRow(label, value) {
+  if (!value) return "";
+  return `<div><b>${label}</b>: ${escapeHtml(String(value))}</div>`;
+}
+
+function escapeHtml(s) {
+  return s.replace(/[&<>"']/g, c => ({ "&": "&amp;", "<": "&lt;", ">": "&gt;", '"': "&quot;", "'": "&#39;" }[c]));
+}
+
+function render(jobs) {
+  if (!jobs.length) {
+    jobsRoot.innerHTML = `<div class="empty">No jobs yet — submit one above to get started.</div>`;
+    return;
+  }
+  jobsRoot.innerHTML = jobs.map(j => {
+    const sCls = statusClass(j.live?.status);
+    const elapsed = fmtElapsed(j.live?.elapsed_s);
+    const isOpen = expanded.has(j.job_id);
+    const logs = (logCache.get(j.job_id) || []).join("\n");
+    const canCancel = sCls === "running" || sCls === "queued";
+    return `
+      <div class="job-row ${isOpen ? "expanded" : ""}" data-id="${j.job_id}">
+        <span class="dot ${sCls}" title="${j.live?.status ?? "unknown"}"></span>
+        <span class="kind">${KIND_ICON[j.kind] || "▶"}</span>
+        <div>
+          <div class="job-title">${escapeHtml(jobTitle(j))}<span class="id">${j.job_id.slice(0, 12)}</span></div>
+          <div class="job-sub">${escapeHtml(jobSubtitle(j))}</div>
+        </div>
+        <div class="job-meta"><span class="status">${(j.live?.status || "unknown")}</span>${elapsed}</div>
+        <span class="chevron">▶</span>
+      </div>
+      <div class="job-detail" data-id="${j.job_id}">
+        <div class="params">
+          ${paramRow("submitted", j.submitted_at)}
+          ${paramRow("requester", j.requester)}
+          ${paramRow("flavor", j.params?.flavor || j.live?.flavor)}
+          ${paramRow("started", j.live?.started_at)}
+          ${paramRow("ended", j.live?.ended_at)}
+          ${j.kind === "analyze" ? paramRow("source", j.params?.source) : ""}
+          ${j.kind === "analyze" ? paramRow("prefix", j.params?.prefix) : ""}
+          ${j.kind === "analyze" ? paramRow("n_traces", j.params?.n_traces) : ""}
+          ${paramRow("output", j.params?.output)}
+        </div>
+        <div class="logs" id="logs-${j.job_id}">${escapeHtml(logs) || "<i style=\\"color:#666\\">no logs yet</i>"}</div>
+        <div class="actions">
+          <button class="secondary" data-act="refresh-logs" data-id="${j.job_id}">Refresh logs</button>
+          ${canCancel ? `<button class="danger" data-act="cancel" data-id="${j.job_id}">Cancel</button>` : ""}
+        </div>
+      </div>
+    `;
+  }).join("");
+}
+
+jobsRoot.addEventListener("click", async (ev) => {
+  const actBtn = ev.target.closest("button[data-act]");
+  if (actBtn) {
+    const id = actBtn.dataset.id;
+    if (actBtn.dataset.act === "refresh-logs") { await refreshLogs(id); return; }
+    if (actBtn.dataset.act === "cancel") {
+      if (!confirm("Cancel job " + id + "?")) return;
+      try {
+        await api(`/api/jobs/${id}/cancel`, { method: "POST" });
+        refreshJobs();
+      } catch (e) { alert("cancel failed: " + e.message); }
+    }
+    return;
+  }
+  const row = ev.target.closest(".job-row");
+  if (!row) return;
+  const id = row.dataset.id;
+  if (expanded.has(id)) expanded.delete(id);
+  else { expanded.add(id); await refreshLogs(id); }
+  refreshJobs(/* keepOpen */ true);
+});
+
+async function refreshLogs(id) {
+  try {
+    const data = await api(`/api/jobs/${id}/logs?tail=200`);
+    logCache.set(id, data.lines || []);
+    const el = document.getElementById("logs-" + id);
+    if (el) {
+      el.textContent = (data.lines || []).join("\n");
+      el.scrollTop = el.scrollHeight;
+    }
+  } catch (e) {
+    const el = document.getElementById("logs-" + id);
+    if (el) el.textContent = "(logs: " + e.message + ")";
+  }
+}
+
+let inFlight = false;
+async function refreshJobs(_keepOpen = false) {
+  if (inFlight) return;
+  inFlight = true;
+  try {
+    const jobs = await api("/api/jobs");
+    render(jobs);
+  } catch (e) {
+    jobsRoot.innerHTML = `<div class="empty" style="color:var(--err)">${escapeHtml(e.message)}</div>`;
+  } finally { inFlight = false; }
+}
+
+// Initial load + polling.
+refreshJobs();
+setInterval(refreshJobs, 5000);
+
+// Refresh logs for expanded rows on a tighter cadence.
+setInterval(() => expanded.forEach(id => refreshLogs(id)), 3000);
+</script>
+</body>
+</html>