Skip to content

Runtime worker: DeerFlow 2.x on GPT-6 via OmniRoute (recipe, round 2) - #427

Closed
seathatflowsinourveins wants to merge 4 commits into
mainfrom
claude/runtime-worker-deerflow-20260927
Closed

seathatflowsinourveins wants to merge 4 commits into
mainfrom
claude/runtime-worker-deerflow-20260927

Conversation

@seathatflowsinourveins

@seathatflowsinourveins seathatflowsinourveins commented Sep 27, 2026 •

Copy link
Copy Markdown
Owner

Runtime-worker recipe: DeerFlow 2.x powered by GPT-6 through the local OmniRoute gateway, on rootless Docker, with MCP tools and the token practice. Built by GPT-6 (astra, max) in two rounds.

SOTA sources

  • bytedance/deer-flow v2.1.0 @ 345f08be00c8a9495079b732a39b46aa9af1584e (archive, backend uv.lock and frontend pnpm-lock pinned by sha256)
  • Grader: UKGovernmentBEIS/inspect_evals v0.22.0 @ 41c72eaa2b807ce43f35fac6f3f399318adc7e9d (GAIA) on Inspect AI 0.3.271 @ c2b63a0b0560f9c3b5b5230365e0a8fe08cd3df0
  • Gateway rules: OmniRoute 3.8.51 routing verdict of 2026-09-27

Status

Draft, recipe only. Nothing is installed, no container is started and no model has run.

  • Round 3 is done. The Claude review (D1-D6), the security review (S2-S11) and the common round-3 requirements (C2-C5) got one GPT-6 repair round (gpt-6-astra, max). The dispositions and 16 fail-first/pass batches are in round3-verification.json and .md.
    • Fixed:
      • Arm-derived base URL and model.
      • Inspect runs with --model none against the unchanged GAIA validation scorer.
      • Native DeerFlow run dispatch (start|wait|result|cancel) with stable ids and distinct exit classes.
      • An internal network plus a fixed gateway forwarder.
      • Fresh E2E state and a Redis password.
      • A 96-package hash lock, installed wheel-only with hashes required.
      • A resolved mount allowlist.
      • Dropped capabilities, no-new-privileges and read-only roots.
      • The Serena dashboard is off.
    • Declined, with a residual recorded (S9b): a blanket non-root override for the gateway, frontend and Redis. It conflicts with host-owned 0700/0600 binds and needs host qualification first.
    • Public web, raw ai-memory and Qdrant access are disabled under the forwarding policy. Public-web GAIA coverage is therefore still unqualified.
  • Arms:
    • Control: cx/gpt-6-astra-max through 127.0.0.1:20128.
    • Engines-on: sharedgw/gpt-6-astra-max through 127.0.0.1:20129. Exactly one slash, and the compression header is sent on engines-on only.
    • Usage is counted once, at each arm's entry gateway, by correlation id.
  • Next: the clean locked install and hardened compose start, host enforcement of the network policy, live routing and effort checks, both arms' GAIA E2E with the unchanged scorer, then receipts.

Checks (head 31efe59c, rebased onto e82e6be7)

  • python3 -m unittest tests.test_runtime_worker_deerflow: 38 tests OK. One upstream-grader test is skipped by design until the grader is installed.
  • The registry tests tests.test_osv_lockfile_coverage, tests.test_blind_checkout and tests.test_workflow_security_coverage pass, 57 tests OK. They fixed the failing validate check.
    • The OSV inventory now lists e2e/requirements.txt and e2e/requirements.lock.
    • blind_checkout classifies the review dispositions fixed and declined as labels, and the round-2 GAIA grader decision as label text, because it states a selection.
  • python3 scripts/validate.py: passed after re-registration per docs/lanes.md.
  • The leak scan of added lines found no personal paths, user names or emails.

🤖 Generated with Claude Code

@seathatflowsinourveins seathatflowsinourveins added the lane:foundation Foundation lane: Claude/Codex setup, hosts, memory, RAG, research, workers label Sep 27, 2026
@seathatflowsinourveins
seathatflowsinourveins force-pushed the claude/runtime-worker-deerflow-20260927 branch from b98a1cd to 31efe59 Compare September 27, 2026 21:14
Scout and others added 4 commits September 27, 2026 17:41
…ned install, gateway FW rules, token MCP wiring, skills, frozen E2E)

Written by GPT-6 (gpt-6-astra, effort max) through the Codex stack-worker profile on the OmniRoute lane from the coordinator's brief; tests red then green; no host install or live run yet. Draft pending the Claude cross-family review, the upstream eval-harness mapping and the skills-lifecycle wiring.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ed resources)

GPT-6 (astra, max) round-2 build: Inspect Evals GAIA v0.22.0 on Inspect AI 0.3.271 as the quality oracle; check.py validates transport only.
Gateway calls omit temperature, carry a stable x-omniroute-session and a fresh
Idempotency-Key per logical call, and accept only configurable GPT-6 max routes.
Loopback-only ports in 3730-3799, owned rw-deerflow- resource names and labels,
literal-name cleanup. Not installed, not run: host E2E and A/B remain open.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…h gateway arms

GPT-6 repair builder (effort max) against the Claude review (D1-D6), the
security review (S2-S11) and the common round-3 requirements (C2-C5):
- arms: control cx/gpt-6-astra-max on 20128; engines-on
  sharedgw/gpt-6-astra-max (one slash) on 20129, with the compression header
  only there; base URL and model derived from the selected arm; max effort,
  no temperature;
- Inspect runs with --model none and worker metadata; the unchanged GAIA
  validation scorer stays the grader;
- native DeerFlow runs dispatch (start/wait/result/cancel) with stable run
  ids, deterministic paths and distinct exit classes;
- security: internal network plus a fixed gateway forwarder, fresh E2E state,
  Redis password, a 96-package hash lock installed wheel-only with hashes, a
  resolved mount allowlist, dropped capabilities, no-new-privileges and
  read-only roots, Serena dashboard off;
- usage counted once at each arm's entry gateway by correlation id.
Declined with a residual: a blanket non-root override for gateway, frontend
and Redis (conflicts with host-owned 0700/0600 binds; needs host
qualification).

Repository registry fixes: the OSV inventory lists e2e/requirements.txt and
e2e/requirements.lock; blind_checkout classifies fixed/declined as labels and
the round-2 GAIA grader decision as label text.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…in for Python 3.12

The validate job runs Python 3.12.3 and failed three DeerFlow tests (run 36350986103):
- dispatch.main used parse_args(). On 3.12 it binds the optional positional
  (prompt, nargs="?") before the options, then rejects a prompt that follows
  them (start --run-id ID PROMPT). parse_intermixed_args() is argparse's
  documented mode for options mixed with positionals (Python 3.7+).
- install_grader creates the grader venv from sys.executable and refuses any
  interpreter but the lock's CPython 3.13, which is correct for a real install.
  Its two tests now stand in that interpreter through the recipe module's own
  sys name. A new test pins the refusal on 3.12 and that nothing runs.

Reproduced with /usr/bin/python3.12: errors=3 before; 40 tests OK after on
Python 3.12.3 and 3.13.15.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@seathatflowsinourveins

Copy link
Copy Markdown
Owner Author

Closing as superseded by #535, which carries every path this PR changed (byte-identical or with later qualification, routing and test updates), per the 2026-10-04 superseded-work cleanup (GPT-lane classification with per-path evidence). The branch stays available for history.

🤖 Generated with Claude Code

@seathatflowsinourveins
seathatflowsinourveins deleted the claude/runtime-worker-deerflow-20260927 branch October 8, 2026 07:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

lane:foundation Foundation lane: Claude/Codex setup, hosts, memory, RAG, research, workers

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant