Active Context
The engagement’s current state, read on demand. .claude/CLAUDE.md § Active Context points here; it moved out of CLAUDE.md so that it is not loaded into every request (BUG-322’s PR). Reconcile it against the injected gate state at session start (.claude/CLAUDE.md § Boundaries, ALWAYS DO), and update it in the change that makes a sentence stale.
Current state only; last reconciled 2026-10-05 (Open risks’ forge-et-iaac PR, uat and Bedrock-quota sentences: 2026-11-19; the critical path, the uploads and mailbox sentences and Open risks’ dev-estate sentences: 2026-10-27; the rest of the cloud bring-up sentences: 2026-10-26; the provider seams: 2026-10-22; the rest of Open risks: 2026-10-16). The dated narrative of how we got here (per-slice histories, meeting fold-ins, decision cascades) lives in .forge/engagement-log.md; a paragraph that stops being current moves there rather than being rewritten here.
Where the engagement stands. Gates 1–3 have passed (Gate 1 with risks, 2026-07-07; Gate 2 with risks and spikes, 2026-08-06; Gate 3, 2026-09-17, verdict Tharinda, re-sliced 2026-10-17, 2026-10-26 and 2026-11-06), and Foundation sealed on 2026-09-16; its two cloud halves (FS-1 Terraform, FS-5 CloudWatch/SNS/SES) are carried by F38, now in progress. The features are in .forge/features.md and tracker.yaml, each with a spec under .forge/specs/ and explicit blocked_by: Phase 1 is in-progress (started 2026-09-17, build target 2026-11-20, milestone 2026-11-27); Phase 2 is locked (build target 2027-01-01, milestone 2027-01-08) — each build target is the tracker’s due_date, one week before the Statement of Work milestone below (Phase 1’s set 2026-09-29, Phase 2’s moved from 2026-12-24 by the lead on 2026-11-08). Work proceeds by /forge-deliver <id> on a Phase 1 feature. Each feature’s status, owner and dates are in tracker.yaml and the dashboard (.forge/dashboard/index.html); this section does not repeat them. The critical path is what is left of the cloud bring-up: F38-e, F38-f and F38-g end in manual checks on dev, each handed to a person to carry out from its run sheets (F38-e wave 3, F38-f wave 2, F38-g wave 4); F38-h, the first UAT environment, has merged its waves 1, 2 and 3 (forge-et-iaac#11 on 2026-11-13), and the uat instance was applied the same day without its backend service (below); wave 4, the deploy-uat workflows, is in progress (forge-et-be#425 for the backend, the forge-et-fe work item after its plan’s re-review), and its waves 5 and 6, the rehearsal and the round’s apply, can start once wave 4 is done and F38-g is done — Empire’s yes of 10 Nov (Kavindu Anthony, progress meeting note §8) is recorded in F38-h spec Rev 8, and the DevOps team’s UAT setup is still owed (.forge/owed.md); F38-i, the production account, is tracked in Phase 1 while its code waves are worked (moved from Phase 2 on 2026-11-02, F38-i spec Rev 3); its production waves wait until production is needed, after Phase 1’s UAT round, so it moves back to Phase 2 before Phase 1 seals unless it is done. The backend has been deployed to dev on every merge to main since 2026-10-09 and the CRM frontend since 2026-10-13, and sign-in on dev works (all built by the DevOps team ahead of the F38-b and F38-d specs); what the Foundation milestone (Fri 16 Oct, not fully met) still lacks is the mailbox proved working on dev (F38-f, F38-g), while the first upload on dev succeeded on 2026-10-26 (F38-d WI-6.1); the AWS resources for both and the scheduler were applied on 2026-10-20 by the delivery team’s F38-b forge-et-iaac#3, and their values are in the backend’s task definition since forge-et-be#300, 2026-10-22; the capture app’s pipeline is in place (F38-d wave 5, forge-et-pwa#91), and its first deploy to dev ran on 2026-10-25.
Commercial frame (binding — L-001). Proposal accepted 2026-08-16: USD 15,500 one-time + USD 360/month support. The AWS account is Empire Teas’ own — held by Kavindu Anthony, paid by Empire Teas’ card and billed to them directly by AWS; SOW ET-002 (Altrium-run, USD 650/month) is superseded and was never signed (.forge/client/aws-sow.md, marked superseded). Running costs are quoted from .forge/client/sow-cost-annexe.md (now an estimate of Empire Teas’ own AWS bill), never the proposal’s §A.2.2/§A.8. Dates, from the revised Statement of Work (SOW-ET-001 v1.1): Foundation incl. the cloud environments and deploy pipeline Fri 16 Oct 2026; Phase 1 Fri 27 Nov; Phase 1 UAT to Thu 10 Dec; Phase 2 (with the capture app, which moved to Phase 1 on 2026-11-11 — § Phase scope; the SOW is not yet amended) Fri 8 Jan 2027; Phase 2 UAT, stabilisation and sign-off Fri 29 Jan 2027 — all assuming the AWS environment is ready by Fri 2 Oct, and moving day for day after it (.forge/client/project-timeline.md). Review windows: five business days for Phase 1 and interim deliveries, ten for Phase 2 and final acceptance. The lead’s rule: finish building one week before each milestone — hence the phase build targets above. Each UAT round runs in a temporary environment; development never doubles as UAT. Where the proposal and an internal document disagree, the proposal wins and the divergence is stated as a post-issue note on its header, never absorbed (.claude/rules/source-trust.md). The SOW-ET-001 v1.1 text lives outside this repo (a Word document). The Phase 1 acceptance criteria are drafted in .forge/client/phase-1-acceptance-criteria.md and not yet issued; still unwritten here: §17 terms and Phase 2’s acceptance criteria.
Team. Four full-stack developers — Tharinda Rajapaksha (lead), Dinil Gamage, Biseni Perera and, from 2026-09-29, Nanndish Satgunarajah — each carrying development, QA and harness work; waves are sliced vertically, no backend/frontend split, and no fixed lanes: each developer takes the next unblocked feature from the tracker (.forge/team-guide.md § Working in Parallel — Four Developers). Pairings and the client-contact rule: .forge/team-guide.md.
Work-item sizing (team — the Gate-3 standing instruction). Decompose into small vertical work items, each verifiable end-to-end across FE + BE and sized for quick review and test; when in doubt, split smaller. Every blocked_by is explicit and parallel versus sequential is classified — the dependency graph is the work-division plan across the four developers. The borderline splits marked ◐ in features.md are the first to merge back if the per-feature overhead bites.
Phase scope. A feature’s phase is its delivery_phase in tracker.yaml; the lead’s phasing intent and how Gate 3 applied it are archived in .forge/engagement-log.md (Active Context snapshot, 2026-09-23). Constraints that still bind: bulk email is committed to Phase 2 as F37-a (the weekly price broadcast) and F37-b (one message to a saved contact list, with opens and clicks per send) — the lead told Empire on 2026-10-09 (PRD Rev 57); F37-a was built early and is accepted with Phase 1 (PRD Rev 100). The proposal does not cover bulk sending (L-001), so the relief is a Phase 2 descope agreed with Empire at or before the Phase 1 hand-over. F37-b is started early in Phase 1 since 2026-11-19 (lead: Tharinda), on the started-early terms later in this paragraph, and the delivery team decided OQ-62 under the rule below on 2026-11-19 (PRD Rev 103: the sender’s own mailbox, everyone on the list unless unsubscribed, bounces read), for Empire to confirm at the hand-over; no other Phase 1 feature depends on it. Migration is leads + accounts only — a fresh start on deals — in two runs, a first load at the start of the parallel run (a date agreed with Empire once F4-d and F4-b are ready) and a final run after the data freeze; each run’s records belong at first to its import owner (the signed-in user who runs it), who hands them out with F32-c, the one bulk action brought into Phase 1 (PRD Rev 47). WhatsApp stays out. Phase 2 was trimmed on 2026-10-11 (PRD Revs 60–65): global search, the AI product-brief assistant, AI similar past quotes, the separate price-lookup page, custom fields created during import, the capture app’s same-day email and its offline queue are out of V1 (the capture app keeps and resends a failed upload instead), three borderline pairs were merged, and F1-c drops the computed lifetime value; each dropped or merged item is recorded in Phase 1 (.forge/engagement-log.md, 2026-10-11). Phase 2 work started early (the lead, 2026-11-06; Gate 3 Run 4, PRD Rev 84): the Phase 2 work that needs neither deals nor the capture app is tracked in Phase 1 — F20-g, F22 and F37-a as they are, and list views, list export, attachments and custom fields on accounts, contacts and leads as F32-d, F23-d, F33-a and F34-c, split from F32-a, F23-a, F33 and F34-a, which keep the deal, quote, sample, standard and brief parts. On 2026-11-11 F20-g, F22 and F23-d moved back to Phase 2, none of them done (F20-g released, its spec and wave-1 draft PRs left open), and F20-k, email block lists, moved from Phase 1 to Phase 2 with them (the lead); Dinil took F20-g back into Phase 1 the same day to finish it, on the same terms as the other started-early items. F37-a (the weekly price broadcast) moved back to Phase 2 the same day, not started; it was later taken back into Phase 1, started early on the same terms (delivery_phase: 1; decomposed 2026-11-15), its two waves merged on 2026-11-17 and it closed on 2026-11-18, and it is accepted with Phase 1 (tracker.yaml F37-a notes). On 2026-11-15 the lead took F23-d back into Phase 1 to start it, on the same terms. On 2026-11-20 the lead took F20-k (email block lists, after F20-f) and F22 (contact sync, after F20-i) back into Phase 1 to start them, on the same terms. Phase 1’s own work, fixes and release still come first (R-ARCH-001); F23-d, F37-b, F20-k and F22, the four still started early and not done, stay in Phase 1 through Empire’s Phase 1 UAT round, so each can go out in drop 1 or a later drop; each one not done by the end of that round moves to Phase 2 (the lead, 2026-11-19). F38-i keeps its own rule. Every started-early feature finished in Phase 1 goes into the Phase 1 acceptance criteria (the lead, 2026-11-18): F20-g (criteria F8 and F9), F37-a (G9) and F32-d (Section N) were added that day, PRD Rev 100, and F23-d, F37-b, F20-k and F22 are each added when done (.forge/owed.md). F34-c was done and added to Phase 1 on 2026-11-11 (the lead): custom fields on accounts, contacts and leads are Section K of the Phase 1 acceptance criteria; F34-a and F34-b stay in Phase 2. F33-a was done and added to Phase 1 on 2026-11-12 (the lead): attachments on accounts, contacts and leads are Section M (PRD Rev 92); F33-b stays in Phase 2. F44 was added to Phase 1 on 2026-11-12 (the lead): a follow-up for a customer email nobody has replied to, asked for at the 10 November meeting, after F20-f (acceptance criterion F7, PRD Rev 92). The capture app (F36-a to F36-d) is Phase 1 scope from 2026-11-11 (the lead): Empire asked at the 10 November progress meeting for it in Phase 1, for its January 2027 trade fairs, and the delivery team and Empire agreed it. It is due with the rest of Phase 1 and, unlike the started-early items, does not move back to Phase 2. The PRD (Rev 91), AD #20 and the client pack, Section L of the Phase 1 acceptance criteria included, say Phase 1 (updated 2026-11-12, for the move of 2026-11-11); the SOW schedule above, the capture app’s production hosting (F38-i; its UAT hosting is built and first deploys with the round’s -rc tag — F38-h spec Rev 12, 2026-11-19), and Empire’s confirmation of the capture-app review on Fri 20 Nov are follow-ups in .forge/owed.md.
Release plan (the lead, 2026-11-18). Phase 1’s UAT round is two drops on the uat instance. Drop 1 is what is ready around Mon 30 Nov. Empire’s start and pace are not known; the plan assumes about two weeks of testing. During that time the delivery team finishes the Phase 1 work still open, at least F20-f, F20-j and F44 (criteria F3, F6 and F7), which the lead wants in before production, and releases it to UAT as drop 2. Phase 1 therefore seals at the end of its UAT round, not before drop 1 (the lead, 2026-11-19): it stays in-progress through the round, release P1 opens early for drop 1, and a UAT defect fixed in drop 2 belongs to Phase 1 (.claude/rules/tracker.md § Delivery Phase Rules). One uat instance serves both drops and Phase 2’s round, and is decommissioned after Phase 2 (AD #23, revised 2026-11-18). Between the rounds it may be paused with its data kept; the lead leans towards stopping the backend and frontends, and Shyamin Ayesh decides. A person does every pause, revive and destroy by hand, never on a date. After drop 1, what Empire raises is triaged and each item is planned into drop 2, a later UAT release or later work; UAT issues usually come before Phase 2 features, and change requests are planned case by case (.claude/rules/tracker.md § UAT issue triage + routing). After UAT, if Empire will use the system, production is created and released, CRM and capture app together, around late December 2026 or early January 2027, before Empire’s first trade fair of 2027, with the real Pipedrive data imported there. Gulfood 2027 is listed for 15–19 March, not January, so which fair comes first is still to be confirmed. UAT’s records are made-up, with test mailboxes and example.com addresses only (F38-h decision 5, AD #34); whether Empire’s testers sign in with their own Microsoft accounts or the test logins is asked of Empire before drop 1, the lead’s one exception to the rule of no questions before UAT. There is no code guard on UAT’s email: the testers and Kavindu are told plainly the risks of connecting their own mailboxes and of loading real Pipedrive data into UAT. The follow-ups are in .forge/owed.md: the two drops and what to tell Empire with drop 1, how UAT sits between the rounds, triaging what Empire raises, telling Kavindu how UAT’s data works, and fixing the go-live date and starting the production-account conversation.
Live decisions to build against (each recorded where it lives): Microsoft-only login onto a pre-created, active user (AD #28, AD #34); access configured, never predefined — no built-in roles, the first admin and an editable IT Admin role seeded by migration per environment, a minimal catalog of view/manage per entity plus a few special keys, with one key per Settings page (AD #41, replacing the single settings.manage), one role per user, at least one active admin — a holder of both the Users and Roles keys — always (AD #34 — built by F5-k; the per-page keys and the two-key rule, AD #41 — built on feature/settings-page-access); one owner plus shares per record, one visibility rule for lists and records, sharing transitive and managed under manage, built once as the F5-f substrate, done before any list read-path (AD #35, .forge/design/data/record-sharing-model.md); no scope on a grant — a key reaches down the reporting line, and an entity’s all key reaches every record (AD #37 — built by F5-l); anyone whose role holds an entity’s manage and who reaches a record, a colleague it was shared with included, may reassign it, share it and remove its shares (AD #35, confirmed by the lead 2026-09-30, PRD Rev 46); every write and security event audited in one trail (AD #36 — built by F39: a writing feature calls the audit service and adds its event codes to forge-et-be/docs/audit/README.md, and a write with no audit event fails the backend test suite); no products catalog — the Buying Standard is the sellable record for bulk and private label alike (PRD Rev 27, FS-15); one price per standard per week, effective-dated rows (PRD Rev 23, AD #14; whether private-label standards are weekly-priced is OQ-61), and that price is a four-figure build-up — LKR → + overhead (LKR) → converted to USD → + margin → final USD, all four on the append-only row, only the final USD figure required (PRD Rev 31, Tharinda 2026-09-17; OQ-43 refined not reversed — USD is still the only published currency and still not a price-key axis; the FX rate is kept on each price row and the overhead is the difference of the two LKR figures, and the workbook is taken as it is today, with no fixed template — PRD Rev 53; the unit of the LKR figures is read as rupees per kg, as built, under the decision below, and OQ-59’s record is settled when work next reaches it); weekly pricing is finished without more input from Empire’s Tea Room (the lead, 2026-11-13): the points never confirmed with them — what a blank workbook cell means, whether prices are set before or after the auction, the unit, their online auction system, and whether two sales can start on one day — are met by keeping the screens simple and flexible as built (Carry Forward All; a round published after its sale has started applies from that day, though the standards it replaces have no price from the sale’s start until then, so the Tea Room guide advises publishing before the sale starts), and nothing more is built on a guess about their practice; a change follows feedback from use or UAT (.forge/engagement-log.md, 2026-11-13 and 2026-11-14); every lead is researched before it is ranked, as a background job on a SearchProvider seam shared with F36 (ranked from the inquiry when research cannot run, and ranked again once its research completes — PRD Rev 83) — built on Exa with no second search provider, chosen by Empire on 2026-11-10 (SP-ARCH-004 closed; PRD Rev 30, .forge/design/ai/lead-research-source-options.md); single-module, feature-first code layout (AD #31); the AI layer on Bedrock behind AIProvider (AD #19), built so a model can be swapped by configuration when Bedrock adds or retires a version — model id in config only, versioned prompt files written to carry across models, an evaluation set per AI feature (.forge/design/ai/ai-provider-strategy.md §4c; every AI feature spec cites it); the CRM’s UI from shadcn/ui — new and changed screens use src/ui/shadcn, and screens built before it are moved over one at a time from .forge/design/ui/shadcn-backfill.md (AD #39). Of the provider seams, MailboxProvider (built by F20-a), AuctionPriceProvider (built by F26-a, its V1 implementation answering “no suggestion”) AIProvider (built by F12, forge-et-be#229) and SearchProvider (built by F11-c, forge-et-be#297, merged 2026-10-22) exist; ErpHandoffService (F19) does not yet — R-FND-001: the first consuming feature builds each on FS-11’s SeamGuard pattern.
No questions to Empire before UAT (the lead, 2026-11-13). Where a spec, plan or design needs an answer from Empire, the delivery team decides it instead, choosing what is simplest for Empire’s staff to understand, the minimum that meets the need, and flexible enough to support the different ways they might work; the decision and its reason are recorded where the question was kept (the spec’s ## Open Questions or decisions, owed.md, the PRD signals), and a change Empire asks for at UAT is then made as ordinary work. A question already waiting on Empire is decided the same way when work reaches it: the owed.md rows titled “Ask Empire …” and the open OQs stay where they are until then, and are read as the delivery team’s decision to make, not a question to send. Weekly pricing was the first area decided this way (the live decision above; .forge/engagement-log.md, 2026-11-13).
Open risks. Dev is partly usable (tracked as F38): as of 2026-10-25 the backend runs at https://api.dev.crm.empire.lk and the CRM frontend at https://dev.crm.empire.lk, in us-east-1, CRM01 can sign in (register §A-5 closed 2026-10-13), SES can send from noreply@dev.crm.empire.lk (production access granted 2026-10-21; the backend has no SES sender yet — F38-e proves a send under its task role; F36-c no longer builds it, because the capture app sends no email (AD #42), so which feature does is .forge/owed.md’s), lead triage (F11-d) and lead research (F11-c) run, the scheduled jobs run, and the alert topic and failed-deploy rule exist (F38-e’s forge-et-iaac#7, applied 2026-10-23; no deploy has failed since to prove the email), and the capture app’s shell is deployed at https://capture.dev.crm.empire.lk (since 2026-10-25, F38-d wave 5), and the first upload on dev succeeded on 2026-10-26 (F38-d WI-6.1), but the mailbox is not yet proved by use; the built estate differs from F38 in ways listed in environments-and-identity.md §4. F38-b’s forge-et-iaac#3 was merged and applied on 2026-10-20 (d230b133, dev.yaml run 39475850266): the KMS keys, the attachments bucket, the mailbox ingest queue, EventBridge Scheduler and the backend’s grants exist, dev.yaml applies only from dev, and the database password’s automatic rotation is off (F38 Rev 14; the dev database stays unencrypted by the lead’s decision of 17 Oct, F38 Rev 17; production’s is encrypted with the AWS-managed aws/rds key, F38-i decision 14; the database fixes are optional in dev, F38 Rev 18; the lead’s dev-only exceptions for the CI key and svc-harness are F38 Rev 9). Those resources’ values are in the backend’s task definition since forge-et-be#300 (2026-10-22, F38-d FR-6), so the mailbox waits only on being walked on dev (F38-f, F38-g); the capture app deploys from its own build workflow (F38-d wave 5, merged 2026-10-25), onto the hosting applied by forge-et-iaac#5 on 2026-10-22; the last forge-et-iaac PRs merged are #13, which lets the backend’s task role call Haiku 5.5, and #14, which gives both frontends a Content-Security-Policy, both applied on 2026-10-28 (the CSP’s signed-in check on dev is owed), then F38-i’s #15, wave 1’s base/ options production needs, each defaulting to dev’s value, and F43’s #16, which lets the task role call both routes of all three models, both applied on 2026-11-04 (#15 with No changes.), then Dinil’s #18 and #19 (F11-g’s lead-ranking queue, 9 Nov; F20-g’s scheduled-email-send schedule, 12 Nov), and F38-h’s #11 (waves 1 and 3: the options for a destroyable instance, overlays/uat and uat.yaml), merged on 2026-11-13 with no change to dev, then #20 to #24 (17–19 Nov: F37-a’s price-broadcast-send schedule, F36-c’s and F36-b’s queues, the capture app’s UAT hosting, the lead-triage timeout); the uat instance exists since 2026-11-13 and runs v0.1.0-rc1 (the backend and the CRM, deployed by deploy-uat on 2026-11-16; CRM01 signed in on 17 Nov; the capture app’s hosting applied on 19 Nov, with no capture-app release tag yet), and its backend refuses the price-broadcast-send job every minute until the next -rc carries F37-a (.forge/owed.md; F38-h’s tracker notes); five PRs are open on 19 Nov — Senuthmi’s #17, overlays/prod, F38-i’s wave 4, reviewed approve and held by Shyamin Ayesh until production is needed, and #25 to #28 (environments-and-identity.md §4, “PRs, as read”); dev’s database had its operating-system update applied on 2026-11-05 (the engagement log of 2026-11-06); and Bedrock serves Haiku 4.5 but refuses Sonnet 5.5 and Haiku 5.5, whose quotas are 0 (register §C-5). AWS has refused every quota request (the latest, two Service Quotas requests for Haiku 4.5 on 17 Nov, closed with no change — environments-and-identity.md §4 item 4), and on 2026-10-26 answered the lead’s appeal: the account keeps new-account limits until it has a few billing cycles of Bedrock usage and spend, with no date, so the 5.5 models stay unusable and Haiku 4.5 allows 10 requests a minute, under which the backend paces a burst of AI calls (F42). The lead decided the same day that Haiku 4.5 runs every AI feature in every environment until AWS lifts the limits, and that the backend paces its Bedrock calls (.forge/design/ai/ai-provider-strategy.md §5 row 10; the pacing is built by F42, and its proof on dev is the register’s §B F42 block). Re-check with git -C ../forge-et-iaac log origin/dev. The Foundation milestone of Fri 16 Oct (not fully met) still lacks the mailbox proved on dev; the mailbox and every feature that uses Sonnet 5.5 or Haiku 5.5 stay unproved on dev until each is available there. R-ARCH-001 (restated 2026-10-12 in tracker.yaml: at the measured pace Phase 1 fits the 20 Nov build target with room, and that room goes first to Phase 1’s own fixes, the feedback from Empire and Altrium stakeholders, regression testing, the release and the cloud work — Phase 2 features are taken into Phase 1’s duration only after all of it is done; Phase 2 does not fit its build target on its own (worked out against 24 Dec; the target moved to Fri 1 Jan 2027 on 2026-11-08, adding the Christmas and New Year week, and the figures were not re-run) — the lead takes the question of which Phase 2 features Empire will actually use to Empire at the Phase 1 hand-over (one meeting with UAT drop 1, about Mon 30 Nov at Empire’s office — the lead, 2026-11-19); the figures count weekends and public holidays but not personal leave or out-of-hours work; the forge-estimate baselines measure agent time without human queue time and cannot stand in for the measured pace); R-FND-001…004 (seams, the six list-screen states owed by the first screen feature, the cloud-gated halves, the backend Slow-lane trend); R-PRD-001/002 architected around. OQ-9 gates F9. Empire gives Pipedrive exports only, never admin access (client/dependencies.md §1.3): the open OQs once tagged to the Pipedrive audit now wait on a deals export for Phase 2 or on questions the lead raises with Empire later (.forge/owed.md).
Owed, with owners: .claude/CLAUDE.md § Active Context keeps the rule; the list is .forge/owed.md.