dev-decide 5.2.0: decision graph and redesigned pages

Install or update in Claude Code:

/plugin marketplace add pawanptu/dev-decide
/plugin install dev-decide@dev-decide
/plugin install dev-decide-graph@dev-decide

Try the web console: https://dev-decide.onrender.com

Added

• Decision graph. http://127.0.0.1:47321/graph shows, per kind of task (type · area), what decided (the
  trained route model, the zero-shot match or a /dev-decide question), which agent or skill was suggested, which
  one Claude really used afterwards (from a new PostToolUse hook on the Agent and Skill tools) and how long each
  step took, with a filterable timeline, a project picker, 1 hour to 30 days, and live refresh. The hint server
  records it after answering, so the hint is no slower.
• dev-decide-graph plugin: the same graph as a side panel, opened with /dev-decide-graph (a text tree in the
  terminal, the graph in the desktop app and VS Code). Needs Claude Code 2.1.310 or later.
• Settings: "decision_graph": {"enabled": true, "keep_days": 30, "prompt_chars": 40}. Records stay on the machine
  in ~/.claude/dev-decide/traces/, never in the repository; of a prompt only the first prompt_chars characters
  are kept, keys and tokens masked. A server listening beyond localhost (serve.py) records nothing.

Changed

• Decision graph page redesigned: a plain-English summary at the top, a "read it left to right" guide, stat
  tiles that count up with sparklines and meters, an animated step-by-step replay of the latest request, a flow graph
  with moving particles, hover tracing, tooltips, keyboard focus and a table view, leaderboards (agents and skills
  used, kinds of tasks, who decided and how fast), and a timeline with per-step timing bars, day groups and filters.
  Live refresh keeps the previous frame on screen and toasts each new request. Colors are the validated categorical
  slots (checked for color-blind separation in light and dark); reduced motion turns every animation off.
• Side panel: the SVG now carries the totals, the latest request, check and cross badges, and moving dots on
  the busiest links; the terminal shows a text tree with bars and check marks.
• Web console rebuilt with Alpine.js (its CSP build, vendored in skill/vendor/, so no eval and no CDN): an
  animated example decision in the hero, a three-step console (what to decide, your input, how to answer) with
  tiles, a staged progress view while it decides, a verdict banner, a "where the answers came from" bar, answers
  ordered by trust with growing confidence meters, history cards, "Try it" on every bundle, and install tabs.
  The page's script now comes from /ui.js, so its CSP is script-src 'self' (no inline script at all).

Fixed

• The console offered Right/Wrong on rough guesses, which the server never logged, so saving one failed.
• On Windows a second hint server could bind the same port (Python's default SO_REUSEADDR), so two sessions
  starting together could split prompts between two servers. It now binds exclusively.

🤖 Generated with Claude Code

dev-decide 5.1.0

5.1.0 (25 October 2026)

Added

• Instant routing hint. A small local server (hookd.py) answers Claude Code's HTTP hook in about 5 ms instead of
  starting Python on every prompt (about 300 ms on Windows). A SessionStart hook starts it when needed. The old
  per-prompt command hook remains available (install.py --router-hook=command).
• Routing by meaning and area. Without a trained route model, matching blends words, the bundled embedding model
  and technical areas (postgres, column, schema count as database). On held-out requests hint precision went from
  80.5% to 88.1%, and false hints with one unrelated skill from 19.5% to 3.2% (tools/eval_router.py).
• Example agents and their route model. Eight subagents in dev-decide-kit/example-agents/ (backend-reviewer,
  db-migrate, frontend-dev, debugger, security-auditor, test-writer, devops-deploy, docs-writer). When they are
  installed, the shipped route model (trained on them) answers routing locally; it is never used for other agents.
• Web page: public demo token filled in for visitors, bring-your-own Anthropic key, one example per decision type,
  right/wrong feedback that the learning loop uses, batch mode (one input per line), history kept in the browser,
  and a warning when the daily Claude budget is nearly used.
• Server: DEV_DECIDE_DEMO_TOKEN (local answers only, nothing about demo requests is stored or learned),
  X-Anthropic-Api-Key (the caller's own key for that request only: never stored, logged or counted),
  DEV_DECIDE_LOCAL_ONLY, a per-client rate limit, a daily model-call cap with an 80% warning, security headers,
  /v1/presets, and richer /health.
• Learning that survives restarts. statesync.py saves what a server learned (models and hashed labels, never
  input text) to a private git repository with a deploy key, and restores it on start.
• GitHub Actions. dev-decide-triage.yml labels new issues and comments; dev-decide-pr.yml rates each pull
  request, keeps one comment up to date and fails the check on block.
• Per-class confidence thresholds. Training measures, per answer class, the confidence its answers need to reach
  the target; weak classes need more before they are answered locally.
• tools/benchmark.py (label, train with calibration, evaluate, write the results table), tools/smoke.py
  end-to-end checks, CI on Linux and Windows with Python 3.8 and 3.12.

Changed

• Triage models retrained with 149 hand-labeled short reports (checked blind by Sonnet) and 772 new generated issues.
  issue_type: short reports never trained on went from 35% to 88% right (96% of its local answers right), 95.3% on
  the held-out test split. severity and domain are now enabled (local answers 94% and 100% right on examples
  neither run trained on). Repositories holding an older starter model get the new ones on their next decision;
  models you trained or the learning loop promoted are never replaced.
• Rule-based guesses no longer call every short report noise.
• The project root is found without starting git (faster on Windows).
• Starter models and a documented settings file (.claude/dev-decide.json) are installed on first use.

Fixed

• Rolling back to the oldest archived model crashed.
• Secrets written as KEY=value without quotes (as in .env files) were not detected; a diff too large to scan
  fully was reported as having no secrets.
• A Claude Code plan or usage limit message was taken for a model answer (and cached by the kit).
• A gate could block on rough local guesses alone; such rules now ask for review, while rules and models still block.
• Reinstalling while the routing-hint server ran failed on Windows (a memory-mapped file); the installer now stops it first.
• Gates passed on answers that were unsure; .pyc files were reviewed as part of a change; a COMMIT -x argument
  could be read as a git option; and the local API answered other websites when no token was set.

Install or update in Claude Code:

/plugin marketplace add pawanptu/dev-decide
/plugin install dev-decide@dev-decide

Full guide: setup_deploy.md. Every number above is reproducible with python3 tools/benchmark.py.

🤖 Generated with Claude Code