R720 agent-team: Plane-2 SDLC pipeline scaffold (pre-deployment) #8

Merged

amoussa1229 merged 7 commits from feature/r720-plane2-scaffold into main

2026-06-17 19:21:27 +00:00

amoussa1229 commented

2026-06-17 18:57:31 +00:00

(Migrated from github.com)

Summary

Scaffolds Plane 2 of the R720 agent-team (design docs/r720-agent-team-design.md, §3.3/§7.1) — the durable, human-gated agentic SDLC pipeline. Pre-deployment only: nothing is provisioned or enabled.

Built under agent-team/:

Foundation: state-store (atomic + integrity), SQLite schema + §3.3.1 compare-and-set, billing-mode seam, Transport ABC, task model.
LangGraph pipeline graph.py (INTAKE→CLARIFY→PLAN, interrupt()/Command(resume), injectable checkpointer), nodes (clarifier/planner/review-loop/builders/verifier), HITL ledger/resume-worker/deadline/recovery/responder, slack/github/claude-code transports, ci_gate, operator_cli, run-team.py, the §3.3.2 split-job CI workflow YAML (authored, disabled), and the P1 sim.

Verified-bug fixes (from the scaffold review, all with regression tests)

builders denylist: 4 execution-proven bypasses (delete, mode-change, copy to, out-of-scope delete) — scan now section-driven off diff --git.
§3.3.1 CAS concurrency: BEGIN IMMEDIATE guarded + retried; per-connection CAS (shared sqlite3.Connection is unsafe); thread-safe db-path derivation; busy_timeout before WAL.
operator CLI: added re-deliver/force-resume; audit-before-mutate; OSError handled.
CI guard: symlink-escape reject, strict-UTF-8 decode, scope-glob canonicalization, egress DEPLOY marker; gate now backed by an extracted-script test.
P1 sim now drives the real graph + SqliteSaver + ledger CAS proving all four §7.1 criteria; fixed a real question_id-stability bug in graph.py.

Tests

564 passed, ruff + format clean. Pinned langgraph-checkpoint-sqlite==3.1.0 (design D9).

NOT in scope (deploy gates remain)

Provisioning IAM/step-ca/Roles Anywhere/confluence-bot, /sh-security-review + GPT-4.1 cross-review sign-off of the CI/IAM surface, live Slack/CI, rsync, live dry-runs, and Adam's approval.

## Summary Scaffolds **Plane 2** of the R720 agent-team (design `docs/r720-agent-team-design.md`, §3.3/§7.1) — the durable, human-gated agentic SDLC pipeline. **Pre-deployment only: nothing is provisioned or enabled.** Built under `agent-team/`: - Foundation: state-store (atomic + integrity), SQLite schema + §3.3.1 compare-and-set, billing-mode seam, `Transport` ABC, task model. - LangGraph pipeline `graph.py` (INTAKE→CLARIFY→PLAN, `interrupt()`/`Command(resume)`, injectable checkpointer), nodes (clarifier/planner/review-loop/builders/verifier), HITL ledger/resume-worker/deadline/recovery/responder, slack/github/claude-code transports, `ci_gate`, `operator_cli`, `run-team.py`, the §3.3.2 split-job CI workflow YAML (authored, disabled), and the P1 sim. ## Verified-bug fixes (from the scaffold review, all with regression tests) - **builders denylist**: 4 execution-proven bypasses (delete, mode-change, `copy to`, out-of-scope delete) — scan now section-driven off `diff --git`. - **§3.3.1 CAS concurrency**: `BEGIN IMMEDIATE` guarded + retried; per-connection CAS (shared `sqlite3.Connection` is unsafe); thread-safe db-path derivation; `busy_timeout` before WAL. - **operator CLI**: added `re-deliver`/`force-resume`; audit-before-mutate; `OSError` handled. - **CI guard**: symlink-escape reject, strict-UTF-8 decode, scope-glob canonicalization, egress `DEPLOY` marker; gate now backed by an extracted-script test. - **P1 sim** now drives the **real** graph + `SqliteSaver` + ledger CAS proving all four §7.1 criteria; fixed a real `question_id`-stability bug in `graph.py`. ## Tests `564 passed`, ruff + format clean. Pinned `langgraph-checkpoint-sqlite==3.1.0` (design D9). ## NOT in scope (deploy gates remain) Provisioning IAM/step-ca/Roles Anywhere/confluence-bot, `/sh-security-review` + GPT-4.1 cross-review sign-off of the CI/IAM surface, live Slack/CI, rsync, live dry-runs, and Adam's approval.

This repo is archived. You cannot comment on pull requests.

No reviewers