Plan · CHG-2026-014880 · Agentic SDLC

change_id: CHG-2026-014880
risk_class: R3
autonomy_tier: A2
controls: ['CHG-04', 'TRC-01', 'HUM-14', 'TPR-05', 'SEC-API-01', 'DP-11', 'FIN-02', 'FRZ-01', 'SOD-01']
data_classification: internal
originator: olaf@freundcloud.com
agent_identity: svc-agent-platform
model_route: gateway/tier-frontier
supersedes: None

Approach

Build the payload first so the gates have something real to act on, then the gates, then the portability layer, then the workflows that run them. Prove each gate by breaking the code it protects and watching it refuse — a gate verified only by passing is not verified.

The repository is built through its own chain: this plan claims every path, so check_plan_conformance.py holds for the scaffolding exactly as it will for a feature.

Files

AGENTS.md
README.md
REVIEW.md
Makefile
pyproject.toml
.gitignore
CODEOWNERS
.github/**
service/**
scripts/**
policy/**
ops/**
.agent/**
docs/**
evidence/.gitkeep

Sequence

Payload service and the closed loop. → verify: make build test lint green.
Policy YAML and the seven gates. → verify: each gate refuses a deliberately broken change and passes once reverted.
Skills, routes, MCP allowlist, OTel collector. → verify: Substitution Test rises.
Eval suite. → verify: 24 cases, --mode static at or above 90%.
Stage 6 detection and its unit tests. → verify: pytest scripts/tests/ green, including the drift case that no threshold would catch.
Seeded artifact chains. → verify: check_artifact_header.py passes, chain complete.
Workflows. → verify: valid YAML, Substitution Test reaches 12/12.
Evidence query and the demo script. → verify: the §6.2 supervisory question answered from the repository in one command.

Risks

Risk	Likelihood	Mitigation
A gate passes vacuously and nobody notices	high — the default failure mode of this kind of work	Every gate has a negative test in `scripts/demo/negative/`; the Substitution Test refuses to award a mark for a document that merely claims a property
Demo evidence is mistaken for real audit records	low, high impact	Labelled as demo output in the README, the query tool and the bundle
The org has no Copilot coding agent, so Stages 2/3/5 cannot run live	medium	Copilot is invoked in one step per stage; the gates and the chain are unaffected, and the fallback is an `@copilot` PR mention
The repository exempts itself from its own gates	medium	This plan; R4 in the spec

Rejected

Building the gates before the service. Faster to start, and every gate would have been verified against a hypothetical rather than real code.
Skipping the negative tests to save time. That is precisely the corner whose absence the whole document argues against.

Tests

service/tests/ — the payload’s own suite
scripts/tests/test_detect_anomaly.py — the Stage 6 detector, including drift
scripts/tests/test_review_tally.py — the severity arithmetic that gates a merge
scripts/demo/negative/ — one scripted branch per gate, each of which must go red

Rollback

The repository is new; rollback is deletion. No downstream consumer exists yet.