loop-spec

v6.7.0

Loop-driven development from a spec file or prose prompt for Codex. Semantic autonomous routing selects micro, bounded debug, compact, or the full seven-phase cycle, then delivers a verified PR.

CodexClaude Code31 Skills

By aztecheadLicense: MIT0 GitHub starsUpdated 6 days ago

Directory evidence

Runtimes: Codex and Claude Code
Parsed components: 31 skill or MCP entries
Source updated: Sep 17, 2026
Manifest status: Canonical path parsed

The directory validates manifest shape and source location. It does not execute the plugin or provide a security endorsement. Review the indexing methodology →

Install loop-spec for Codex and Claude Code

Installs for the current user

codex plugin marketplace add aztechead/loop-spec
codex plugin marketplace upgrade loop-spec-marketplace
codex plugin add loop-spec@loop-spec-marketplace

Paste and run these commands in a terminal with Codex. They add and refresh the loop-spec-marketplace catalog, then install this plugin.

Compatibility: the page URL and API slug “loop-spec” remain stable.

Codex: loop-spec@agent-plugin-marketplace → loop-spec@loop-spec-marketplace

The installer fetches third-party code from the source repository shown on this page. This directory validates manifest structure and source location, but does not perform a security audit; review the manifest, components, and source before installing.

Get the source manually

$ git clone https://github.com/aztechead/loop-spec

Clone the source repository, then follow its setup instructions to add the plugin to a compatible client. The repository root is the plugin root.

Plugin files

loop-spec/

├── .codex-plugin/plugin.json
├── .claude-plugin/plugin.json
├── skills/assess/SKILL.md
├── skills/auto/SKILL.md
├── skills/checking-gates/SKILL.md
├── skills/cycle/SKILL.md
├── skills/debug/SKILL.md
├── skills/deliver/SKILL.md
├── skills/discuss/SKILL.md
├── skills/execute/SKILL.md
├── skills/forensics/SKILL.md
├── skills/intake/SKILL.md
├── skills/iterate/SKILL.md
├── skills/loop-runner/SKILL.md
├── skills/micro/SKILL.md
├── skills/onboard/SKILL.md
├── skills/oneshot/SKILL.md
├── skills/pause/SKILL.md
├── skills/plan/SKILL.md
├── skills/quality-loop/SKILL.md
├── skills/retro/SKILL.md
├── skills/revise/SKILL.md
├── skills/rollback/SKILL.md
├── skills/rules/SKILL.md
├── skills/sentinel/SKILL.md
├── skills/settings/SKILL.md
├── skills/spec/SKILL.md
├── skills/spec-lite/SKILL.md
├── skills/specifying-gates/SKILL.md
├── skills/status/SKILL.md
├── skills/verify/SKILL.md
├── skills/walkthrough/SKILL.md
└── skills/watch/SKILL.md

Included Skills31

assessskills/assess/SKILL.md

Use when asked to assess codebase health or review fragile files across repositories. Writes ASSESSMENT.md without changing source code or committing. Use /loop-spec:debug for a specific failure.

autoskills/auto/SKILL.md

Route autonomous script or SDK requests to micro, debug, compact, or full based on repository evidence. Use when the caller has not selected a route. Do not replace a skill the user named in an interactive session.

checking-gatesskills/checking-gates/SKILL.md

Verify a user-gate task when the user or an enabled hook requests it. Report evidence from the specified check. Use /loop-spec:specifying-gates for unclear methods and /loop-spec:verify for the full VERIFY phase.

cycleskills/cycle/SKILL.md

Use when starting or resuming repository work from a feature description or spec file. Runs SPEC -> DISCUSS -> PLAN -> EXECUTE -> VERIFY -> ITERATE -> DELIVER, one phase per invocation. Use /loop-spec:debug for stack traces and /loop-spec:micro for one-file fixes.

debugskills/debug/SKILL.md

Use when the user pastes an error, stack trace, failing test, or a vague "something's wrong / flaky / slow". Writes docs/loop-spec/debug/{slug}/BUG.md (triage, red repro, fix, sibling sweep, verify). Do not use for a new feature (that's /loop-spec:cycle) or a tiny unrelated edit (that's /loop-spec:micro).

deliverskills/deliver/SKILL.md

Push the verified SHA, update its PR, check required CI results, and mark the PR ready after ITERATE converges. Internal phase of /loop-spec:cycle at currentPhase=deliver.

discussskills/discuss/SKILL.md

Resolve design choices in SPEC.md and run the selected challenger review. Internal phase of /loop-spec:cycle. Start there for repository work.

executeskills/execute/SKILL.md

Run PLAN tasks with the execution mode selected from task dependencies and available tools. Internal phase of /loop-spec:cycle. Start there for repository work.

forensicsskills/forensics/SKILL.md

Use when a loop-spec feature is stuck or failed and the user says "what happened", "why is this stuck", or "diagnose this run". Writes .loop-spec/forensics/ and does not change the feature. Do not use to resume or fix the run (that's /loop-spec:cycle) or to debug product code (that's /loop-spec:debug).

intakeskills/intake/SKILL.md

Use when converting pasted text or a local file into a spec draft before starting the cycle. Pass existing SPEC.md files to /loop-spec:cycle and stack traces to /loop-spec:debug.

iterateskills/iterate/SKILL.md

Judge the integrated result against the original goal. Route unresolved gaps through the graph or advance to DELIVER. Internal phase of /loop-spec:cycle. Start there for repository work.

loop-runnerskills/loop-runner/SKILL.md

Compile specs/goals into autonomous Claude Code, OpenCode, Google ADK, or Codex loops and run them safely. Use whenever the user wants a supported harness hands-off — "implement this spec", "keep going until tests pass", "babysit my PRs", overnight/cron/CI, agent fleets, ralph or /goal-style loops, or worries about runaway agents. Do not use for an interactive SPEC interview (that's /loop-spec:cycle) or a single small edit (that's /loop-spec:micro).

microskills/micro/SKILL.md

Use when completing a small task with criteria, verification, and PR delivery. Also supports on, off, and status. Use /loop-spec:cycle for features needing a spec and /loop-spec:debug for unexplained failures.

onboardskills/onboard/SKILL.md

Use when the user says "set up loop-spec", "onboard", or "configure grill/discipline/workflow". Writes .loop-spec/grill.conf, discipline.conf, workflow.json, and a RULES.md seed. Do not use this to start a feature cycle (that's /loop-spec:cycle); everything it configures can also be set by hand.

oneshotskills/oneshot/SKILL.md

Implement, review, and verify a change with at most three files in its SPEC footprint. Internal cycle phase selected by lib/graph/probes/oneshot.sh. Start at /loop-spec:cycle.

pauseskills/pause/SKILL.md

Use when the user says "pause", "stop for now", "pick this up later", or needs a handoff. Writes HANDOFF.json and .continue-here.md on the active feature. Do not use to cancel a cycle, to resume (re-invoke /loop-spec:cycle), or to roll back code (that's /loop-spec:rollback).

planskills/plan/SKILL.md

Create PATTERNS.md, PLAN.md, and tasks.json from the frozen SPEC. Internal phase of /loop-spec:cycle. Start there for repository work.

quality-loopskills/quality-loop/SKILL.md

Use when reviewing modified files or named paths before commit. Fix findings when authorized, within the review limits. Does not commit. Use /loop-spec:verify inside a cycle.

retroskills/retro/SKILL.md

Use when finding repeated patterns in loop-spec telemetry. Report writes RETRO.md; apply adds rule candidates. Use /loop-spec:forensics to investigate one run.

reviseskills/revise/SKILL.md

Use when the user says "address the review comments", "revise this PR", or pastes a loop-spec PR number/URL after human review. Fixes execute-class items on the PR branch. Do not use to start new work (that's /loop-spec:cycle) or to ingest a ticket that is not PR review (that's /loop-spec:intake).

rollbackskills/rollback/SKILL.md

Use when the user says "rollback", "restore the checkpoint", or names a loop-spec-checkpoint-* tag. History-safe checkout, then a new commit. Do not use git reset --hard, and do not use this to pause (that's /loop-spec:pause).

rulesskills/rules/SKILL.md

Use when the user says "add a rule", "list the rules", or "show RULES.md". Reads and writes .loop-spec/RULES.md (add/list/render/path). Do not use this to implement a feature; prefer a deterministic --check over a prose note.

sentinelskills/sentinel/SKILL.md

Use when scanning work sources into the sentinel queue or running its eligible work through autonomous cycles. Use /loop-spec:cycle for a specific task and /loop-spec:assess for codebase health.

settingsskills/settings/SKILL.md

Use when the user says "grill on/off", "discipline on/off", "simplicity lite/full/ultra/off", "human-code on/off/probe", or "loop-spec settings". Writes the per-project .loop-spec/<mode>.conf files the SessionStart hooks read. Do not use this to implement a feature; it only flips switches.

specskills/spec/SKILL.md

Write SPEC.md from repository evidence and recorded decisions. Resolve intent questions before approval. Internal phase of /loop-spec:cycle. Start there for repository work.

spec-liteskills/spec-lite/SKILL.md

Inspect the requested change and record its footprint. Fill the driver's short spec or continue to the full SPEC instructions. Internal cycle entry for NEXT phase=spec. Start at /loop-spec:cycle.

specifying-gatesskills/specifying-gates/SKILL.md

Use when a user-gate task needs a clear verification method. Ask for criteria, proof, scope, and failure policy, then update task metadata. Does not run verification.

statusskills/status/SKILL.md

Use when the user says "status", "what's running", "stats", "metrics", or "trust level". Read-only over feature.json, events.jsonl, result.json, and the metrics contract. Do not use this to start, pause, or deliver a feature.

verifyskills/verify/SKILL.md

Check acceptance criteria and code review, record evidence, and route failures for remediation. Internal phase of /loop-spec:cycle. Start there for repository work.

walkthroughskills/walkthrough/SKILL.md

Use when handing a branch, PR, or feature to a human reviewer, or when the user says "walk me through this change". Writes REVIEW-ORDER.md. Do not use to implement the change or to address review comments (that's /loop-spec:revise).

watchskills/watch/SKILL.md

Use after a cycle PR merges, when cron/CI should watch the default branch for the watch window. Appends the verdict to the feature's committed run digest. Do not use this to reopen or continue a cycle — it never starts one.

Plugin manifests2

.codex-plugin/plugin.json

{
  "name": "loop-spec",
  "version": "6.7.0",
  "description": "Loop-driven development from a spec file or prose prompt for Codex. Semantic autonomous routing selects micro, bounded debug, compact, or the full seven-phase cycle, then delivers a verified PR.",
  "author": {
    "name": "aztechead",
    "email": "[email protected]",
    "url": "https://github.com/aztechead"
  },
  "homepage": "https://github.com/aztechead/loop-spec",
  "repository": "https://github.com/aztechead/loop-spec",
  "license": "MIT",
  "keywords": [
    "spec-driven",
    "skills",
    "tdd",
    "agents",
    "codex"
  ],
  "skills": "./skills/",
  "hooks": "./hooks/codex-hooks.json",
  "interface": {
    "displayName": "loop-spec",
    "shortDescription": "Spec-driven cycle through ready-for-review PR",
    "longDescription": "Give Codex a feature description or a pre-authored spec. loop-spec runs SPEC, DISCUSS, PLAN, EXECUTE, VERIFY, ITERATE, and DELIVER from one source tree shared with Claude Code, OpenCode, and Google ADK.",
    "developerName": "aztechead",
    "category": "Productivity",
    "capabilities": [
      "Read",
      "Write"
    ],
    "websiteURL": "https://github.com/aztechead/loop-spec",
    "defaultPrompt": [
      "Run the loop-spec cycle: $loop-spec-cycle <description>",
      "Run loop-spec autonomously: $loop-spec-auto <description>"
    ]
  }
}

.claude-plugin/plugin.json

{
  "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
  "name": "loop-spec",
  "description": "Loop-driven development from a spec file or prose prompt for Claude Code. Semantic autonomous routing selects micro, bounded debug, compact, or the full seven-phase cycle, then delivers a verified PR.",
  "version": "6.7.0",
  "author": {
    "name": "aztechead",
    "email": "[email protected]"
  },
  "homepage": "https://github.com/aztechead/loop-spec",
  "repository": "https://github.com/aztechead/loop-spec",
  "license": "MIT",
  "keywords": [
    "spec-driven",
    "skills",
    "tdd",
    "agents",
    "claude-code",
    "pi",
    "opencode"
  ],
  "outputStyles": "./output-styles/"
}

Browse the “spec-driven” tag →

by SuperClaude Team

Transform Claude Code into a structured development platform with 29 /sc: commands, 23 specialized agents, 7 behavioral modes, and MCP server integration

Claude Code

55Updated 2 months ago

agentic-workflow

by Gabriel Trabanco

Stack-agnostic agent skills for a disciplined, doc-driven feature/issue workflow: design, plan, execute by phases, review, audit, and ship — with a full internal review pack and merge gates.

Claude Code40 Skills

21Updated 2 days ago

scaffolding

by komluk

13 agents, 38 skills, 19 commands, 17 hooks — spec-driven, multi-agent orchestration for Claude Code with per-phase model tiers, opt-in lifecycle hooks, and optional cross-device semantic memory.

Claude Code38 Skills

15Updated 3 days ago

For maintainers

If you maintain this plugin, link to this source-backed listing from your README so users can review its manifest and indexed components.

[loop-spec on Agent Plugins Marketplace](https://pluginsmp.com/plugins/loop-spec)