title: Sprint 2 design harness for API/MM-Cron and MWMC — who does what, prompt queues, and the week before (25–31 Oct) status: accepted direction (Atul, 24 Oct: Sonnet expedites API/MM-Cron and MWMC alongside the harness); queues live in docs/prompts/ author: Claude (chat, MMDEP-PM) date: 2026-10-24 related: TD-025, TD-032, TD-033 (technical-decisions register); plans/direct-agents-design-harness-horizons §B; roadmap v4
Sprint 2 design harness: API/MM-Cron and MWMC¶
Sprint 2 runs 1–14 Nov. This week (25–31 Oct) we run the design harness once by hand on both projects, so sprint 2 starts with designs, ready issues and agents that know their queue. Five workers, each in its lane:
| Worker | Role | Lane / scope | Where it writes |
|---|---|---|---|
| Claude (MMDEP-PM) | Designs the queues (this document). Reviews high-risk LLDs (API write-route enforcement, NAV de-duplication on production). Keeps the decision register | design, decisions | MMDEP |
| API agent A (current ChatGPT project) | Lane P: production and high-risk API work | api.moneymarche.com, security, contracts, production data fixes | GitLab api / mm-crons, branches into production only with Atul's approval |
| API agent B (older ChatGPT project, resumed; API-2) | Lane S: staging, data load, MR clean-up, OpenAPI, MM-Cron logging and inventory, low-risk code | stgapi, api-phase1-development, mm-crons staging | GitLab api / mm-crons, lane:staging |
| MWMC agent (ChatGPT) | Page types, test plans, SEO audit triage, section inventory with WPDev | website and webapp tests | GitLab sandbox / webapp, MMDEP relay |
| Sonnet (Claude Code) | Builds the tools both projects need, in MMDEP, running on the laptop runner against staging and production read-only: NAV data-quality checker, contract tests, Lighthouse CI, crawler checks. No GitLab write is needed | MMDEP tools/ and tests/, jobs register |
MMDEP (merge by risk) |
| Local code harness | Low-risk issues with routing blocks: tests, docs, small fixes, in api lane S and MMDEP | for:code, risk low |
GitLab api (B8 source), MMDEP |
Atul: one PROCEED per agent per day at most; business questions only, through for-atul (TD-024).
The loop for each project (TD-025)¶
- The queue file
docs/prompts/<lane>/S2/QUEUE.mdlists numbered steps. Each step gives: owner, inputs, outputs (exact paths), a check, and the harness job that follows. - On PROCEED, the agent takes the first step whose dependencies are done, delivers it, and records
runs/<step>.mdin its repo (or the MMDEP relay for GitLab agents without a repo path). Then it continues with the next step, until it is blocked or the queue is empty. - After each step: CI checks it, and the post-step job (B40) or Sonnet's tools act on it. For example, an LLD with a work-items table becomes ready issues (
mmdep design issues, B36), which the harness picks up.
Week of 25–31 Oct (design-harness rehearsal)¶
| Day | API agent A (lane P) | API agent B (lane S) | MWMC | Sonnet | Harness |
|---|---|---|---|---|---|
| Sat–Sun | — | Atul resumes it with the API-2 prompt | reads the SEO audit (relayed by inbox-relay) |
PROCEED: unstarve the harness, B42 model roles, then S2 tools | idle until issues are ready |
| Mon–Tue | P1: write-route security LLD → enforcement plan with a work-items table | S1: MR clean-up (23 api + 22 mm-crons → at most 5 each) | M1: page-type classification of the URL inventory | T1: NAV data-quality checker; T2: Lighthouse CI job | runs ready issues from S1 and M1 |
| Wed–Thu | P2: NAV duplicates on production: root cause and de-dup design (high risk, Claude reviews) | S2: staging data load plan, run step 1 | M2: SEO audit → issues and crawler checks | T3: contract tests for routes used by MWMC/PE/WPDev; T4: crawler dead-end/CTA/SEO checks | issues from P1/M2 work items |
| Fri | P3: /api/health on production; /category flags fix ready for release | S3: fix staging index/nav 404 and nfos 500; OpenAPI draft | M3: test plan per page type, with the section inventory from WPDev | T5: sprint metrics in the sprint page | — |
Done by 31 Oct: - every S2 story has an LLD or an issue list; - at least 10 ready low-risk issues sit in the harness queue; - the four Sonnet tools are merged and run daily.
Sprint 2 stories (1–14 Nov) for these two projects¶
| Story | Owner | Value (roadmap v4) |
|---|---|---|
| API write routes secured on staging, consumers on keys; production enforced with Atul's approval | A (P), B (S) | risk reduction (security) |
| NAV duplicates and gaps fixed, with a daily checker green for 7 days | A (P) design, B (S) staging, Sonnet checker | data quality, PE and website trust |
| Staging loaded from cleansed production data; stgapi routes match production | B (S) | unblocks staging tests |
| OpenAPI for catalogue and NAV routes; contract tests | B (S), Sonnet | MCRM, PE, WPDev integration |
| MM-Cron logging and direct-DB inventory; fund details for all AMCs (TER, size, inception) | B (S) | R1 fund pages |
| MWMC: page-type tests, Lighthouse budgets, SEO audit fixes tracked, dead-end/CTA checks daily | MWMC, Sonnet | R1: impressions and minutes |
| First fund factsheet page from the production API | MWMC + A (P) | R1 |
Measured on the sprint page: stories done, harness-merged PRs, first-pass CI rate, steps per PROCEED, Atul-minutes.