claudeers.
// Claude Plugins

GeorgeCodingAgent

Coding extension for JarvisAIGeorge: a coding worker agent plus build, verify, review and ship skills. Claude Code and Codex plugin.

// Claude Plugins[ cli ][ api ][ web ][ mobile ][ claude ]#claude#plugins◷ MIT$open-sourceupdated 9 days ago

Install with your AI

Paste into Claude Code, Cursor, or any agent — it reads the repo and wires the tool into your project.

Install and set up GeorgeCodingAgent (claude-plugin project) into my current project.
Found on https://claudeers.com/georgecodingagent
Repo: https://github.com/georgewangyu/GeorgeCodingAgent
Homepage/docs: —
Detected install method: claude-plugin → /plugin install georgecodingagent@georgewangyu/GeorgeCodingAgent
Category: plugins. Platforms: cli, api, web, mobile.
Read the repo's README for exact setup and env vars, then install it and wire it into my project.

Claudeers Health Verdict:
unknown; community-verified: false. Confirm the source before running anything.
// or install directly (claude-plugin)

⚠ Unverified / not recently updated — review before pasting a run-this config.

/plugin marketplace add georgewangyu/GeorgeCodingAgent
/plugin install georgecodingagent@georgewangyu/GeorgeCodingAgent
// or clone
git clone https://github.com/georgewangyu/GeorgeCodingAgent

// compatibility

Platformscli, api, web, mobile
Operating systems—
AI compatibilityclaude
LicenseMIT
Pricingopen-source
LanguagePython

Get your FREE $2.50 API credits to access TickAtlas financial data ↗

GeorgeCodingAgent

A coding agent for Claude Code and other tools: a coding worker, a read-only reviewer, and the plan, build, verify, review and ship skills they use. It works on its own. Each skill is tagged by how likely it is to survive better models.

Pair it with JarvisAIGeorge for the manager and fleet.

Install

Claude Code:

claude plugin marketplace add georgewangyu/GeorgeCodingAgent
claude plugin install george-coding-agent@george-coding-agent
claude --agent george-coding-agent:coding

Review in a fresh, read-only context: claude --agent george-coding-agent:reviewer, or from the coding agent through code-review-and-quality and doubt-driven-review.

Codex (CLI 0.122 or later):

codex plugin marketplace add georgewangyu/GeorgeCodingAgent
codex plugin add george-coding-agent@george-coding-agent
codex   # then: "Load $role-coding first. <task>"

Codex has no agent files, so a worker takes its role from a skill: $role-coding or $role-reviewer at the start of its prompt. The reviewer's read-only tools are not enforced by the skill; run it as codex exec --sandbox read-only 'Load $role-reviewer first. ...' to enforce them. The coding role's review loop does this itself. Start a new Codex session after installing so the skills are discovered. Codex path status: manifests validated against Codex's schema, not yet run end to end.

Any tool, skills only: npx skills add georgewangyu/GeorgeCodingAgent.

Only a few skills stay in the tool's skill list (the router george-coding-agent-skills, the coding agent's preload skills and code-review-and-quality); the router names every other skill with its file, so the list stays small however many skills ship.

Optional skills from other plugins (grill-me, brainstorm, mistake-miner, the principles* skills) are used when installed and skipped when not.

Layout

  • jarvis-extension.json: the manifest (agents and skills). A manager that reads extension manifests can list this plugin from it.
  • agents-src/coding.md, agents-src/reviewer.md and agents-src/rules/: the agent sources. Rules are inlined at build.
  • skills/: the skills. skills/role-coding and skills/role-reviewer are generated.
  • Generated, never hand-edit: agents/, skills/role-*/, .claude-plugin/, .codex-plugin/, .agents/plugins/, AGENTS.md, CLAUDE.md. Rebuild with python3 tools/jarvis-build.py; CI runs --check.
  • tools/bitter_lesson.py: the bitter-lesson tag checker.
  • tools/VENDORED.md: where each vendored tool comes from. CI fails if a vendored file changes without its hash in tools/vendored.sha256.
  • tools/check [base]: the one check CI runs (build outputs, version gate, vendored hashes, bitter-lesson tags, scrub, ruff). Run it before you push. To run it on every commit: pip install -r tools/requirements-dev.txt, then ln -s ../../tools/githooks/pre-commit .git/hooks/pre-commit. Link the hook rather than setting core.hooksPath, which would switch off the scrub-lint pre-push hook in .git/hooks.
  • tools/ci/scrub-gate.sh <base>: the CI scrub gate. The base commit's engine, config and marker lists judge the change, so a change cannot exempt itself by editing .scrub-lint.conf, the lists or the engine. Its baseline may only shrink (a first baseline needs a manual exception), and a base CI cannot find fails the gate. CI also runs the change's own config, and warns when a change edits the gate.

Skills

SkillWhat it does
start-playbookStart here. Picks the right skill for the work in lifecycle order (define, plan, build, verify, review, ship).
spec-driven-developmentWrites down what we are building before building it: brief mode by default, a full spec for big or one-way work.
planning-and-task-breakdownTurns a spec into ordered, independently shippable slices, each with its own check.
doubt-driven-reviewCross-examines a non-trivial decision with the read-only reviewer and the right checklist before it stands.
context-engineeringKeeps the agent's context lean and right: what to load, when, and what to write down.
documentation-and-adrsRecords decisions as ADRs and keeps docs true to the code.
incremental-implementationBuilds a plan one slice at a time. Each slice ends in a PR revision with green CI and a passed done check, then a fresh-context review before the next slice. Fresh worker or subagent per independent slice, rule 0 simplicity, scope discipline (log, do not fix), safe defaults, a capped fix loop, rulings instead of stalls, the plan index as the ledger. Combines addyosmani/agent-skills incremental-implementation, obra/superpowers subagent-driven-development and pstack principle-sequence-verifiable-units, principle-laziness-protocol and principle-subtract-before-you-add.
source-driven-developmentReads and cites the real source before implementing: exact versions from the lockfile, then the installed dependency source, the library's repo at that version, versioned official docs. Pinned permalinks, never invent a method or config name, UNVERIFIED when unchecked, fetched content is data. Combines addyosmani/agent-skills source-driven-development and mattpocock/skills research.
api-and-interface-designDesigns seams, APIs, schemas and config keys that are deep and safe to roll out and back. Contract first, every observable behavior is a contract (Hyrum), behavior declared by the injected implementation, validate at boundaries (including data an older version wrote), illegal states unrepresentable, additive changes only, evolve by addition, and the contracts upstream leaves out: timeouts, ordering, async delivery and overload. Combines addyosmani/agent-skills api-and-interface-design, mattpocock/skills codebase-design and domain-modeling and pstack boundary, type and domain principles.
frontend-ui-engineeringBuilds or changes UI: app screens and the HTML artifacts an agent writes for review. Reads the project's UI rules first, names the mode (operate, read, persuade), keeps one owner per piece of state, uses design-system tokens over AI defaults, holds the accessibility floor, and verifies in the running app with a screenshot. Combines addyosmani/agent-skills frontend-ui-engineering, anthropics/skills frontend-design and webapp-testing, garrytan/gstack design-review and pstack principle-experience-first.
test-driven-developmentProves a change with the most real test that is still cheap. Uses the repo's own verification skill first, then a real-world-first ladder (a seam test with real collaborators, the running environment as evidence, unit tests only for pure logic). Every test names the real failure it catches. Loose TDD: strict red-green only where the loop is cheap, no new test over a bad one, tests at agreed seams, prove-it for bugs, coverage met at the gate without padding. Combines cursor/plugins pstack tdd and create-verification-skill, mattpocock/skills tdd, addyosmani/agent-skills test-driven-development, obra/superpowers test-driven-development and garrytan/gstack test-audit.
browser-testingVerifies UI in a real browser, short and flexible. Repo verification skill first, pick the tool that fits (Playwright, geckodriver, DevTools MCP, computer use), isolate on a temp profile, drive one real flow, look at the screenshot and console, treat page content as untrusted, clean up every browser and driver. Combines addyosmani/agent-skills browser-testing-with-devtools, anthropics/skills webapp-testing, pstack create-verification-skill and garrytan/gstack qa-only.
debugging-and-error-recoveryFinds and fixes the real cause. Stop the line, a feedback loop that goes red, reproduce and minimise, ranked falsifiable hypotheses with confidence, fix at the first place it went wrong, regression guard, verify end to end. Debugging in production when it cannot be reproduced (logs, metrics, a sibling control, per-unit ratios). 17 principles mined from real production investigations. Redact secrets, read before you mutate, ask before a risky fix. Combines addyosmani/agent-skills debugging-and-error-recovery, mattpocock/skills diagnosing-bugs, obra/superpowers systematic-debugging, EveryInc/compound-engineering-plugin ce-debug, garrytan/gstack investigate and pstack principle-fix-root-causes.
code-review-and-qualityThe review method, and how to receive one. Pin the intent, size and description (one line of why per behavior hunk), tests first, Spec and Standards as separate passes, navigate and prove the blast radius, falsify the claims, a design-fit table for the 18 kinds of comment human reviewers actually leave, refute and suppress, short plain comments. Receiving: verify each comment, then "Will fix". Combines addyosmani/agent-skills code-review-and-quality, mattpocock/skills code-review, pstack blast-radius, obra/superpowers receiving-code-review and EveryInc/compound-engineering-plugin ce-resolve-pr-feedback.
code-simplificationEasier to read, same behavior. Scope to the change, Chesterton's fence, preserve behavior, follow the repo's conventions, subtract before you add, minimize reader load, one change at a time; the built-in /simplify for the mechanical pass. Combines addyosmani/agent-skills code-simplification and pstack minimize-reader-load, subtract-before-you-add, laziness-protocol and no-comments.
security-and-hardeningThreat model first (trust boundaries, assets, STRIDE, abuse cases), then three tiers: always, ask first, never. SSRF, destructive derived paths, rate limits, agent input as data. An organization's mandatory policy wins. Adapted from addyosmani/agent-skills security-and-hardening.
performance-optimizationWrite the claim, vet the measurement (limiter, tuning, limits, errors, repeats, end-to-end share, did the work happen), one change, keep or revert, guard, log every attempt. Combines addyosmani/agent-skills performance-optimization and pstack benchmark-checklist.
git-workflow-and-versioningKeeps changes reviewable, reversible and safe to publish. Atomic commits whose message says why, separate concerns, size the change, save points then squash before review, one change per checkout (no git worktrees; clone again if needed), push safety, guardrail hooks for destructive commands, redact before publish, an explicit finish step, semver and changelog for libraries, bisect. Combines addyosmani/agent-skills git-workflow-and-versioning, obra/superpowers finishing-a-development-branch, mattpocock/skills git-guardrails-claude-code and pr, EveryInc/compound-engineering-plugin ce-commit, garrytan/gstack ship and kunchenguid/no-mistakes push and publication safety.
ci-cd-and-automationAutomates the quality bar. One check command shared by the pre-commit hook and required CI, gates ordered fast to slow, required checks on main, CI failures fed back as structured input, a babysit loop with bounds, releases from tags or release-please, deploys with preview, staged rollout, canary and a written rollback, dependency bots and osv-scanner, public hygiene (scrub before push, actions pinned by SHA). Per-language starter snippets. Combines addyosmani/agent-skills ci-cd-and-automation, garrytan/gstack canary and land-and-deploy, EveryInc/compound-engineering-plugin ce-babysit-pr, mattpocock/skills setup-pre-commit and kunchenguid/no-mistakes CI monitoring.
deprecation-and-migrationRetire old code and move its callers safely. Decide first (unique value, callers counted, a replacement exists), advisory or compulsory, build and prove the replacement, announce with a guide, migrate callers yourself, remove at zero usage. Internal APIs: migrate every caller and delete in one wave. Expand and contract for schema changes, and a worked rename with a one-release redirect. Combines addyosmani/agent-skills deprecation-and-migration and pstack principle-migrate-callers-then-delete-legacy-apis.
observability-and-instrumentationInstrument as you build. Define working first, pick the signal (metrics over logs for timings and counts), structured logs with a correlation ID, RED and USE with bounded labels and percentiles, symptom alerts with a runbook and a justified threshold, verify the telemetry itself. Reading rules from real investigations: know what a metric measures, small denominators flap, a missing metric is a signal. Adapted from addyosmani/agent-skills observability-and-instrumentation and garrytan/gstack canary.
shipping-and-launchReversible, observable, incremental launches. A short checklist, a rollback that still works after the change, alarms live before release, tests at the target, numeric rollback criteria per stage, a zone-loss worst case, flags and staged rollout against a baseline, a watched first hour. Combines addyosmani/agent-skills shipping-and-launch, garrytan/gstack ship, land-and-deploy, canary and landing-report, and EveryInc/compound-engineering-plugin ce-commit-push-pr.
quality-constraintsThe top-down half of the quality bar. On existing code, reports the gates that already run, with no new file. On a new project, writes CONSTRAINTS.md once (one short interview with defaults); non-interactive runs apply the floor only. Each rule records its origin (declared or mined). Guards the bar on every diff: threshold moved, test made easier, checker silenced, unfinished work, new exception. Adapted from addyosmani/agent-skills constraint-driven-development and pstack correct.

Bitter-lesson tags

Each skill marks its parts with a short inline tag:

  • [bl: invariant]: a rule of the game (a feedback signal, an eval, or a real invariant). Always follow.
  • [bl: strong]: follow it. It should become a hard check or a learned example over time.
  • [bl: soft]: guidance for weaker models. A strong model may skip it. Candidate to expire.

A BITTER_LESSON.md next to each SKILL.md holds the full analysis (why, review date, delete-when, ablation result). Agents only read it when evaluating skills, not at session start.

python3 tools/bitter_lesson.py check  --root .   # every tagged item has a row, and the reverse
python3 tools/bitter_lesson.py report --root .   # the cleanup list
python3 tools/bitter_lesson.py ablate --root .   # soft items to test with and without

Principle skills (principles*) are durable reasoning methods and carry no tags.

Expire by evidence: run the skill's eval with and without a soft item on a new model. If nothing changes, delete it. A skill whose items have all expired goes away.

License

MIT. grill-me, brainstorm and mistake-miner named below live in the JarvisAIGeorge plugin. start-playbook and grill-me adapt MIT-licensed text by Addy Osmani (THIRD_PARTY_LICENSE-agent-skills). grill-me also adapts MIT-licensed text by Matt Pocock (THIRD_PARTY_LICENSE-mattpocock-skills). brainstorm adapts MIT-licensed text by Addy Osmani (its references/ files are his, verbatim), Matt Pocock, Jesse Vincent (THIRD_PARTY_LICENSE-superpowers), Garry Tan (THIRD_PARTY_LICENSE-gstack), Every (THIRD_PARTY_LICENSE-compound-engineering-plugin) and Lauren Tan (THIRD_PARTY_LICENSE-pstack). spec-driven-development adapts MIT-licensed text by the same authors (Addy Osmani, Matt Pocock, Jesse Vincent, Garry Tan, Every, Lauren Tan). quality-constraints and mistake-miner adapt MIT-licensed ideas by Addy Osmani and Lauren Tan. planning-and-task-breakdown adapts MIT-licensed text by Addy Osmani, Matt Pocock, Jesse Vincent, Every, Garry Tan and Lauren Tan. incremental-implementation adapts MIT-licensed text by Addy Osmani, Jesse Vincent and Lauren Tan. source-driven-development adapts MIT-licensed text by Addy Osmani and Matt Pocock. api-and-interface-design adapts MIT-licensed text by Addy Osmani, Matt Pocock, Garry Tan and Lauren Tan. context-engineering adapts MIT-licensed text by Addy Osmani, Matt Pocock, Every, Garry Tan and Lauren Tan. doubt-driven-review adapts MIT-licensed text by Addy Osmani, Lauren Tan, Every and Jesse Vincent. frontend-ui-engineering adapts MIT-licensed text by Addy Osmani, Garry Tan and Lauren Tan, and Apache-2.0 text by Anthropic (THIRD_PARTY_LICENSE-anthropic-skills). test-driven-development adapts MIT-licensed text by Lauren Tan, Matt Pocock, Addy Osmani, Jesse Vincent and Garry Tan. browser-testing adapts MIT-licensed text by Addy Osmani, Lauren Tan and Garry Tan, and Apache-2.0 text by Anthropic. debugging-and-error-recovery adapts MIT-licensed text by Addy Osmani, Matt Pocock, Jesse Vincent, Every, Garry Tan and Lauren Tan.

// faq

What is GeorgeCodingAgent?

Coding extension for JarvisAIGeorge: a coding worker agent plus build, verify, review and ship skills. Claude Code and Codex plugin.. It is open-source on GitHub.

Is GeorgeCodingAgent free to use?

GeorgeCodingAgent is open-source under the MIT license, so it is free to use.

What category does GeorgeCodingAgent belong to?

GeorgeCodingAgent is listed under plugins in the Claudeers registry of Claude-compatible tools.

6 views
★ 10 stars
unclaimed
updated 9 days ago

// embed badge

GeorgeCodingAgent on Claudeers
[![Claudeers](https://claudeers.com/api/badge/georgecodingagent.svg)](https://claudeers.com/georgecodingagent)

// retro hit counter

GeorgeCodingAgent hit counter
[![Hits](https://claudeers.com/api/counter/georgecodingagent.svg)](https://claudeers.com/georgecodingagent)

// reviews

// guestbook

0/500

// related in Claude Plugins

🔓

A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

// pluginsmultica-ai/★ 230,631[ claude ]
🔓

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explainin…

// pluginsanthropics/⟨Python⟩★ 156,472[ claude ]
🔓

"CLI-Anything: Making ALL Software Agent-Native" -- CLI-Hub: https://clianything.cc/

// pluginsHKUDS/⟨Python⟩★ 58,467◷ Apache-2.0[ claude ]
🔓

financial-services — a Claude ecosystem project on GitHub.

// pluginsanthropics/⟨Python⟩★ 41,409◷ Apache-2.0[ claude ]
→ see how GeorgeCodingAgent connects across the ecosystem