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
Problem
The Agentic SDLC Playbook v1.0 describes a seven-stage lifecycle, five planes, a portable artifact chain and an autonomy matrix. It is a good document and it is only a document. When we take it to a bank, the CISO asks “what does the gate actually do when someone tries that”, and the honest answer today is “here is the paragraph about it”.
A prospect cannot tell the difference between a control that is designed and a control that operates. Neither, in the end, can a supervisor.
Who is affected
Anyone presenting the playbook — every engagement conversation. The prospect’s CISO, head of engineering and internal audit, who need to agree on what “the agent is allowed to do” means before a programme can start. Our own delivery teams, who otherwise rebuild this scaffolding per engagement.
Success criteria
- A consultant can show a gate refusing a change, live, in under a minute.
- Every claim in the playbook that can be executed, is executed — the Substitution Test scores itself from the repository rather than being self-assessed.
- The repository is built through its own artifact chain. If we exempt ourselves, the demo is a lie the audience will notice.
- Swapping the agent runtime is an edit to one workflow step, not to the operating model.
- Every generated artifact states plainly that it is a demo.
Out of scope
- A real model gateway. Routes are declared and the eval runner speaks to a gateway, but no gateway is stood up here.
- A real deployment target. The release workflow attests and gates; it deploys to
echo. - Anything that would make fabricated evidence look like a real institution’s records.
Constraints
- Must run in a plain GitHub repository with no paid add-on beyond Copilot.
- Must not name a model anywhere, because the Substitution Test checks for it.
- The demo must be reproducible: the same numbers every time it is run.
Open questions
- Does the target org have the Copilot coding agent enabled? Affects Stages 2, 3 and 5.