feat: autonomous agent org — roles, peer review, intake, live demo by bashebr · Pull Request #2 · bashebr/ai-native-sdlc

bashebr · 2026-08-28T17:56:06Z

What

Adds the autonomous agent org capability to ai-native-sdlc: scaffold named, role-based agents into any project so the SDLC loop runs with peer review and less human steering.

init_org.py scaffolds org/ — org chart, role cards (CEO-human, CTO, PM, product engineering agent, engineers, reviewer), reporting/peer-review/escalation protocol, status + review queue, multi-channel intake.
sync_issues.py — GitHub issue intake: pull open issues into org/intake/github/ (idempotent, state-tracked) and push feature tickets.
Human involvement stays at the critical gates: PM first review of intents (CEO on ambiguity), CEO on unresolved spec/plan disagreement, and CEO approval for PR merge and release.
Expense-tracker example is now a real static app (HTML/CSS/vanilla JS, localStorage) with no API calls and no secrets — no backend, no keys, no external requests.

Live demo

https://bashebr.github.io/ai-native-sdlc/ — hosted on GitHub Pages from the gh-pages branch (the app also runs locally and is Vercel-ready via vercel.json).

How to use

python3 skills/ai-native-sdlc/scripts/init_workflow.py my-project --name "My idea" --git
python3 skills/ai-native-sdlc/scripts/init_org.py my-project

See skills/ai-native-sdlc/references/org.md and docs/superpowers/specs/2026-08-28-agent-org-design.md.

Validation

quick_validate.py: 80 passed, 0 failed
test_gate.sh: 25 passed, 0 failed
test_init.sh / test_init_org.sh: pass
unittest discover -s tests: 23 tests OK
Secret scan: no API keys/tokens/credentials in the tree, git history, gh-pages branch, or served HTML.

…trasts

…xamples

- production-gate.sh: match deploy actions (kubectl/helm/terraform/etc.) instead of the word 'deploy'; read-only allowlist ends doc-read false positives; RELEASE_APPROVAL_EXPIRY (ISO-8601 or epoch, fail-closed); jq-independent JSON parsing for hook stdin mode - quick_validate.py: self-contained skill/plugin validator (frontmatter, plugin.json sync, YAML/JSON parse, bash -n, gate suite, scaffold smoke) - run_evals.py: eval-suite runner for Phase 4 with JSON eval format, --min-pass-rate gating, --record history; agent-evals.yml.example now calls it (previously referenced a nonexistent check.sh) - detect_bands.py: Phase 6 control-band reference implementation (rolling-30d mean/sigma, Western Electric rules, drift rule) - init_workflow.py: scaffold workflow-graph.yaml, .gitignore, evals/README; add --dry-run, --git, --name validation - New templates: runbooks/rollback-deploy.md (fixes dangling bands.yaml route), incident.md, PULL_REQUEST_TEMPLATE.md, evals.example.json, project workflow-graph.yaml - tests/ (gate 21 cases, init, band detector, eval runner) and .github/workflows/self-check.yml so the repo practices what it preaches - SKILL.md: version field, rule->enforcement compliance matrix, self-test; README/AGENTS.md validation instructions point at real artifacts; CHANGELOG.md, SECURITY.md

- gate_ledger.py: record/list/verify a hash-chained gates/ledger.jsonl (every gate decision is a committed, tamper-evident record); verify --require-committed rejects uncommitted ledgers; --graph --require-gates enforces completeness against workflow-graph.yaml; duplicate ids are rejected (append-only) - production-gate.sh: RELEASE_APPROVAL=ledger:<id> verified against the ledger (committed, approved, unexpired) before a deploy is allowed; tampered/uncommitted/unknown records BLOCK - Scaffold copies gate_ledger.py, run_evals.py, detect_bands.py, and gates/README.md into new projects - tests: 6 ledger unit tests (chain, tamper, expiry, decision, committed, graph completeness) + 4 gate integration cases (uncommitted blocks, valid allows, unknown blocks, tampered blocks) - quick_validate, SKILL.md (matrix + assets/scripts), README, CHANGELOG updated

GitHub Actions runners have no user.name/user.email configured, so the temp-repo commits in test_gate_ledger.py (test_require_committed) and the ledger section of test_gate.sh failed with exit 128, breaking the self-check job. Set a local identity in both fixtures. Verified by running the full suite with an empty HOME (simulating the runner's missing git config): gate, init, 19 unit tests, quick_validate 51/51 all green.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 40a7881049

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-08-28T18:00:13Z

+        max_seen = max([last_seen] + [int(i["number"]) for i in new_issues])
+        state[key] = max_seen


Page intake before advancing its high-water mark

gh issue list --help describes --limit as the “Maximum number of issues to fetch,” so a repo/filter with more than 100 open issues is only partially read here. If that page contains issue #500 but omits an older #200, this records 500 as the state; all later runs discard #200 because it is no longer greater than last_seen, permanently losing that demand from intake. Paginate through the initial result set or track processed issue IDs/files instead of advancing a numeric high-water mark from a partial page.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-28T18:00:13Z

+
+    try:
+        if args.command == "pull":
+            return _pull(Path(args.config), args.dry_run)


Resolve intake paths from the scaffolded project

Following the documented adoption command in references/org.md:94 from the parent directory leaves this default path relative to that parent rather than to my-project. The configured my-project/org/intake/config.json is therefore ignored; _load_json falls back to {}, and a real pull targets the current GitHub repository and writes ./org/... outside the scaffolded project. Anchor the default config and its relative state/output paths to the script's project root, or require cd my-project before invoking the script.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-28T18:00:13Z

+      const imported = JSON.parse(String(reader.result));
+      if (!Array.isArray(imported)) throw new Error("not an array");
+      state.expenses = [...state.expenses, ...imported];
+      save();


Validate imported records before persisting them

A syntactically valid JSON array is accepted without validating its entries. For example, importing [{}] appends and saves the object before render() throws on e.date.startsWith; the catch only shows an alert, so the malformed data remains in localStorage. On the next reload, load() returns that saved array and the initial render throws uncaught, leaving the demo unusable until the user manually clears site storage.

Useful? React with 👍 / 👎.

bashebr added 22 commits August 22, 2026 22:34

feat: AI-native SDLC skill, plugin, and templates

04d2c0e

docs: make intent.md dual-format requirement explicit

039b4f8

docs: capture traditional vs AI-native design handoff

a64da6e

docs: mirror AI-native design definition

b4f32ab

docs: add traditional build contrast

fda278e

docs: mirror AI-native build definition

bf6abf6

docs: add traditional test contrast

678cdfa

docs: mirror AI-native test definition

df56175

docs: add traditional deploy contrast

8a65251

docs: complete shifts table, governance principles, and remaining con…

f318d6a

…trasts

docs: add framework mapping and clarify scaffold vs workflow artifacts

65c9468

docs: subagents are named, visible, and accountable

1fdc39f

feat: graph engineering reference, workflow-graph asset, and worked e…

816ccac

…xamples

fix: address pre-publish review findings

b23046a

docs: add repo banner and learn-more links to README

ce271c8

docs: add contribution guide to README

f0d36e5

feat: scaffold the autonomous agent org

b17d0f6

feat: GitHub issue intake for the agent org

af74e9b

feat: expense-tracker static demo and docs

03a8026

chatgpt-codex-connector Bot reviewed Aug 28, 2026

View reviewed changes

docs: point expense-tracker demo at live GitHub Pages URL

be76d77

bashebr closed this Aug 28, 2026

bashebr force-pushed the feat/agent-org branch from a7b06c3 to be76d77 Compare August 28, 2026 18:23

bashebr mentioned this pull request Aug 28, 2026

feat: autonomous agent org — roles, peer review, intake, live demo #3

Merged

Navigation Menu

feat: autonomous agent org — roles, peer review, intake, live demo - #2

feat: autonomous agent org — roles, peer review, intake, live demo#2
bashebr wants to merge 23 commits into
mainfrom
feat/agent-org

bashebr commented Aug 28, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Aug 28, 2026

Uh oh!

chatgpt-codex-connector Bot Aug 28, 2026

Uh oh!

chatgpt-codex-connector Bot Aug 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		max_seen = max([last_seen] + [int(i["number"]) for i in new_issues])
		state[key] = max_seen

Conversation

bashebr commented Aug 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What

Live demo

How to use

Validation

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Aug 28, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Aug 28, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Aug 28, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

bashebr commented Aug 28, 2026 •

edited

Loading