feat: initial release — Agent Discipline Standard v0.1.0

The standard (19 criteria / 6 categories), the in-agent audit skill, clean-room reference configs, a worked before/after example, and full community scaffolding (CONTRIBUTING, Code of Conduct, SECURITY, issue/PR templates, CI). Scores verified reproducible across independent blind auditors (0-1 point spread).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: SpinaBuilds

.claude-plugin/marketplace.json

@@ -0,0 +1,14 @@
+{
+  "name": "agent-discipline",
+  "description": "The Agent Discipline Standard — audit the operating discipline of a coding agent's config, from inside the agent.",
+  "owner": {
+    "name": "Antonio Spina"
+  },
+  "plugins": [
+    {
+      "name": "agent-discipline",
+      "source": "./",
+      "description": "Audit the operating discipline of a coding agent's config against the Agent Discipline Standard."
+    }
+  ]
+}

.claude-plugin/plugin.json

@@ -0,0 +1,11 @@
+{
+  "name": "agent-discipline",
+  "description": "Audit the operating discipline of a coding agent's config against the Agent Discipline Standard — scored 0-100, run from inside the agent itself.",
+  "version": "0.1.0",
+  "author": {
+    "name": "Antonio Spina"
+  },
+  "homepage": "https://github.com/SpinaBuilds/agent-discipline",
+  "repository": "https://github.com/SpinaBuilds/agent-discipline",
+  "license": "MIT"
+}

.editorconfig

@@ -0,0 +1,12 @@
+root = true
+
+[*]
+charset = utf-8
+end_of_line = lf
+insert_final_newline = true
+trim_trailing_whitespace = true
+indent_style = space
+indent_size = 2
+
+[*.md]
+trim_trailing_whitespace = false

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,21 @@
+---
+name: Bug report
+about: Something in the standard, skill, or templates is wrong or misbehaves
+title: "[bug] "
+labels: bug
+---
+
+**What happened**
+A clear description of the problem.
+
+**Where**
+Which file / criterion / skill step (e.g. `STANDARD.md` AD-4, `SKILL.md` Step 2).
+
+**Expected vs actual**
+What you expected the audit/standard to do, and what it did instead. If it's a scoring issue, include the config you audited and the score you got.
+
+**Reproducibility**
+If it's a scoring inconsistency: did it vary across runs? By how much? (The standard targets a few-points spread max.)
+
+**Environment**
+Claude Code version, OS, install method (plugin / standalone).

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,5 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Question or discussion
+    url: https://github.com/SpinaBuilds/agent-discipline/discussions
+    about: Ask a question or discuss the standard before filing an issue.

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,21 @@
+---
+name: Feature request / new criterion
+about: Propose a new criterion, dimension, or improvement
+title: "[proposal] "
+labels: enhancement
+---
+
+**What discipline gap does this close?**
+Describe the engineering practice that the standard currently fails to measure.
+
+**Proposed criterion**
+- **Presence test** (a near-deterministic "is this explicit rule present?" yes/no):
+- **What good looks like** (quality signal):
+- **Red flags**:
+- **Suggested category and weight** (remember: weights must still sum to 100):
+
+**Behavioral (`AD-*`) or Enforcement (`EN-*`)?**
+Is this about a rule being *written* (behavioral) or *mechanically imposed by the harness* (enforcement)?
+
+**Why it's not syntactic**
+Confirm this isn't already covered by a deterministic linter (agnix/cclint) — those checks are intentionally out of scope.

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,23 @@
+## What this changes
+
+Brief description, and the issue it closes (e.g. `Closes #12`).
+
+## Type
+
+- [ ] Standard change (criterion / weight / scoring) — **changes everyone's score**
+- [ ] Skill change
+- [ ] Reference config / example
+- [ ] Docs / meta
+
+## Checklist
+
+- [ ] Weights still sum to **100** at both category and criterion level (if `STANDARD.md` changed)
+- [ ] Presence tests stayed near-deterministic; only quality carries judgment
+- [ ] The skill still *reads* `STANDARD.md` — criteria are not duplicated into it
+- [ ] JSON is valid and internal links resolve (CI checks both)
+- [ ] For a standard change: I sanity-checked reproducibility (a re-score lands within a few points)
+- [ ] Conventional commit messages
+
+## Scope confirmation
+
+This PR does one thing. If it grew beyond its title, I split it.

.github/workflows/validate.yml

@@ -0,0 +1,52 @@
+name: validate
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  validate:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: JSON files are valid
+        run: |
+          python3 - <<'PY'
+          import json, sys
+          ok = True
+          for f in ['.claude-plugin/plugin.json', '.claude-plugin/marketplace.json', 'templates/settings.reference.json']:
+              try:
+                  json.load(open(f)); print('OK  ', f)
+              except Exception as e:
+                  print('FAIL', f, e); ok = False
+          sys.exit(0 if ok else 1)
+          PY
+
+      - name: STANDARD weights sum to 100 (both levels)
+        run: |
+          python3 - <<'PY'
+          import re, sys
+          txt = open('STANDARD.md').read()
+          cats = [int(w) for w in re.findall(r'^### .*weight (\d+)', txt, re.M)]
+          crit = [int(w) for w in re.findall(r'^#### .*weight (\d+)', txt, re.M)]
+          print('categories sum =', sum(cats), '| criteria sum =', sum(crit))
+          sys.exit(0 if sum(cats) == 100 and sum(crit) == 100 else 1)
+          PY
+
+      - name: Internal markdown links resolve
+        run: |
+          python3 - <<'PY'
+          import re, os, glob, sys
+          errs = []
+          for md in glob.glob('**/*.md', recursive=True):
+              d = os.path.dirname(md)
+              for m in re.findall(r'\]\((\.[^)\s]+)\)', open(md).read()):
+                  p = m.split('#')[0]
+                  if p and not os.path.exists(os.path.normpath(os.path.join(d, p))):
+                      errs.append(f'{md} -> {m}')
+          for e in errs: print('BROKEN', e)
+          print('broken links:', len(errs))
+          sys.exit(0 if not errs else 1)
+          PY

.gitignore

@@ -0,0 +1,30 @@
+# Secrets — never commit (this repo's own AD-7 / AD-14)
+.env
+.env.*
+*.pem
+*.key
+credentials.json
+
+# Dependencies
+node_modules/
+.pnp
+.pnp.js
+
+# Build / cache
+dist/
+build/
+.cache/
+*.log
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Editors
+.idea/
+.vscode/
+*.swp
+*~
+
+# Local plugin test artifacts
+/tmp/

CHANGELOG.md

@@ -0,0 +1,21 @@
+# Changelog
+
+All notable changes to this project are documented here. The format follows
+[Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to
+[Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.1.0] — 2026-06-11
+
+Initial release.
+
+### Added
+- **The Agent Discipline Standard** ([`STANDARD.md`](./STANDARD.md)) — 19 criteria across 6 categories (Failure discipline, Verification, Security, Delegation & autonomy, Improvement loop, Enforcement), weights summing to 100. Two-part scoring: near-deterministic presence + bounded quality for behavioral criteria; fully deterministic checklist for enforcement criteria.
+- **Audit skill** (`skills/audit-setup/SKILL.md`) — runs inside Claude Code as `/agent-discipline:audit-setup`, reads the standard rather than duplicating it, and emits a 0–100 score with per-criterion evidence and concrete diffs.
+- **Plugin packaging** — `.claude-plugin/plugin.json` + `marketplace.json` for `/plugin marketplace add`.
+- **Clean-room reference configs** — `templates/CLAUDE.reference.md`, `templates/AGENTS.reference.md`, `templates/settings.reference.json` (score in the 🟢 band; verified 97/97/98 across three blind auditors).
+- **Worked example** — `examples/before-after/` (a reasonable-looking config that scores 0/100, with the diffs to fix it).
+
+[Unreleased]: https://github.com/SpinaBuilds/agent-discipline/compare/v0.1.0...HEAD
+[0.1.0]: https://github.com/SpinaBuilds/agent-discipline/releases/tag/v0.1.0