knowledge/planner-memory.md

@@ -1,6 +1,6 @@
 <!-- summarized-through-run: 5 -->
 # Planner Memory
-Last updated: 2026-04-12
+Last updated: 2026-04-14
 
 ## Run history
 - Run 1 (2026-04-05): First planning run. Established prerequisite tree from VISION.md.
@@ -9,26 +9,28 @@ Last updated: 2026-04-12
 - Run 4 (2026-04-11): Foundation+ DONE. Adoption architect DONE (#689, #690). Ops migration DONE (#688). Filed #697.
 - Run 5 (2026-04-12): Dev pipeline reliability DONE (#693, #702 closed). Chat 8/10 (#710, #713 closed). Found architect false-completion bug → filed #735. 0 predictions.
 - Run 6 (2026-04-12): #735 CLOSED (architect false-completion bug). #737 NEW tech-debt (close_vision_issue error handling). Relabeled #707, #712 blocked→backlog for ci_exhausted retry. Added priority+backlog to #737. 0 predictions. 0 new issues filed.
+- Run 7 (2026-04-14): #737 CLOSED, #741 CLOSED. Architect decomposed #426 into sprint (4 sub-issues #745-#748). Labeled all sub-issues backlog. Created vault pending item for SSH key. #749 new bug (dev-poll post-crash deadlock, in-progress). #742 CI exhaustion investigation in-progress. #707/#712 re-blocked pending #742. 0 predictions. 0 new issues filed.
 
 ## Current constraint focus
-1. **Architect error handling** (#737) — backlog, priority, tech-debt. close_vision_issue() swallows PATCH errors → vision issues can stay stuck-open. Small fix, one file. Dev-agent should pick up.
+1. **Website observability sprint** (#426) — architect decomposed. #747 (evidence dir, priority) → #745 (formula) → #746 (format guard). #748 (SSH docs) parallel. SSH key blocked-on-vault. First factory observable in progress.
-2. **Website observability** (#426) — vision, priority. disinto.ai has no analytics. Blocks Adoption landing page AND Ship engagement measurement. Architect should decompose now that #735 is fixed.
+2. **Dev-poll post-crash deadlock** (#749) — in-progress, dev-bot. Self-assigned issues not recovered after container crash. Same class as #741 (startup state vs Forgejo state).
-3. **Example project** (#697) — vision, priority. Needs architect decomposition. Architect unblocked by #735 fix.
+3. **CI exhaustion investigation** (#742) — in-progress, dev-qwen. Blocks #707/#712 (last 2 chat sub-issues for #623).
-4. **Chat ci_exhausted** (#707, #712) — relabeled backlog for retry. #712 has stale PR #733. Last 2 of 10 sub-issues for #623.
+4. **Example project** (#697) — vision, priority. Awaiting architect decomposition. Architect should pick up next run.
-5. **Deploy profiles** — not started. Vision-level. Blocks Ship milestone.
+5. **Deploy profiles** — not started. Vision-level. Blocks Ship milestone. Do not file until Adoption shows more progress.
 
 ## Patterns observed
 - Foundation completed in ~10 days, Foundation+ in ~3 days, dev pipeline reliability in ~1 day. Execution velocity increasing.
-- Chat assistant (#623) rapid progress: 6/10 → 8/10 sub-issues in one day. Final 2 are ci_exhausted — CI reliability is the bottleneck.
+- **Architect lifecycle fully operational**: decomposed #426 into clean sprint with 4 sub-issues. First successful end-to-end architect cycle since bug fixes (#716-#718, #735, #737).
-- Architect lifecycle: 5 bugs found and fixed in one day (#716-#718, #735). #737 is a follow-on from #735's fix PR review. Architect is now functional — should be decomposing #426 and #697 on next run.
+- Cold-boot resilience is a recurring theme: #741 (credential race), #749 (self-assigned deadlock). Both are startup-state-vs-Forgejo-state reconciliation bugs.
-- No predictor activity (0 predictions in runs 4-6). Either predictor is idle or all patterns are accounted for.
+- Chat assistant (#623) 8/10 done. Final 2 (#707, #712) blocked on CI investigation (#742). If #742 resolves, chat could complete quickly.
-- No observables exist — all addressables still lack measurement. Fundamental gap unchanged.
+- No predictor activity (0 predictions in runs 4-7). Stable system or quiet predictor.
-- Graph healthy: 206 nodes, 271 edges, no cycles, no bottlenecks.
+- Graph grew: 206→213 nodes, 271→289 edges. No cycles, no bottlenecks.
+- **First observable in progress** — #426 decomposition is the first concrete step toward measurement. Fundamental gap being addressed.
 
 ## Strategic direction
-- Adoption is the active milestone. 5 of 8 Adoption objectives DONE.
+- Adoption is the active milestone. 6 of 8 Adoption objectives DONE (Architect lifecycle now DONE with #737 closed).
-- **Critical path**: architect decomposes #426 and #697 (now unblocked by #735 fix) → Adoption objectives unblock.
+- **Critical path**: #747 → #745 → #746 (website observability chain) + vault SSH key provisioning. This is the path to the factory's first observable.
-- #737 is a small reliability fix — should be quick for dev-agent.
+- #697 needs architect decomposition — should happen on next architect run.
-- Chat (#623) is 80% done. #707/#712 relabeled for retry — if they fail again, need CI investigation.
+- #749 and #742 are active reliability work — let them complete.
-- Do NOT file Ship/Learn issues until Adoption objectives show progress.
+- Do NOT file Ship/Learn issues until website observability is wired (#426 chain completes).
-- RESOURCES.md and portfolio.md remain empty — populate when addressables materialize.
+- RESOURCES.md and portfolio.md remain empty — populate when first observable materializes.

prerequisites.md

@@ -1,5 +1,5 @@
 # Prerequisite Tree
-<!-- Last updated: 2026-04-12 -->
+<!-- Last updated: 2026-04-14 -->
 
 ## Objective: Foundation — Core agent loop (dev → CI → review → merge)
 - [x] dev-agent picks up backlog issues (dev/dev-agent.sh exists)
@@ -18,7 +18,7 @@ Status: DONE
 ## Objective: Foundation — Planner gap analysis against vision
 - [x] Planner formula exists (run-planner.toml v4)
 - [x] planner-run.sh cron wrapper exists
-- [x] Planning runs established (runs 1–6)
+- [x] Planning runs established (runs 1–7)
 Status: DONE
 
 ## Objective: Foundation — Multi-project support
@@ -29,7 +29,7 @@ Status: DONE
 ## Objective: Foundation — Knowledge graph for structural defect detection
 - [x] networkx package installed in agents container (#220 — closed)
 - [x] build-graph.py exists in lib/
-- [x] Graph report generating successfully (206 nodes, 271 edges as of 2026-04-12)
+- [x] Graph report generating successfully (213 nodes, 289 edges as of 2026-04-14)
 Status: DONE
 
 ## Objective: Foundation — Predictor-planner adversarial feedback loop
@@ -47,6 +47,7 @@ Status: DONE
 - [x] Agent profile/journal fixes (#651, #652, #663 — closed)
 - [x] Cron intervals configured (#682 — closed)
 - [x] Remove external flock (#647 — closed)
+- [x] Credential helper race on cold boot (#741 — closed)
 Status: DONE
 
 ## --- FOUNDATION+ MILESTONE: DONE ---
@@ -68,13 +69,19 @@ Status: DONE
 
 ## Objective: Adoption — Landing page communicating value proposition
 - [x] Website addressable exists (disinto.ai)
-- [ ] Website observability — no engagement measurement (#426 — vision, priority)
+- [ ] Website observability (#426 — vision, priority, architect-decomposed into sprint)
-Status: BLOCKED — no evidence process connected to website
+  - [x] Architect decomposition complete (sprint: website-observability-wire-up, ops PR #10)
+  - [ ] Evidence/engagement directory setup + ops-setup wiring (#747 — backlog, priority)
+  - [ ] Collect-engagement formula + container script (#745 — backlog, depends on #747)
+  - [ ] Format-detection guard in collect-engagement.sh (#746 — backlog, depends on #745)
+  - [ ] Rent-a-human SSH key setup docs (#748 — backlog)
+  - [ ] SSH key provisioned on Caddy host — blocked-on-vault ($OPS_REPO_ROOT/vault/pending/disinto-caddy-ssh-key.md)
+Status: IN PROGRESS — architect decomposed, sub-issues labeled, SSH key needs human provisioning
 
 ## Objective: Adoption — Example project demonstrating full lifecycle
 - [x] Bootstrap path unblocked (#688 — closed)
-- [ ] Example project design and implementation (#697 — vision, priority)
+- [ ] Example project design and implementation (#697 — vision, priority, awaiting architect decomposition)
-Status: BLOCKED — #697 awaiting architect decomposition (unblocked now that #735 is fixed)
+Status: BLOCKED — #697 awaiting architect decomposition
 
 ## Objective: Adoption — Architect lifecycle completeness
 - [x] Architect closes parent vision issues when sprint sub-issues complete (#689 — closed)
@@ -83,13 +90,14 @@ Status: BLOCKED — #697 awaiting architect decomposition (unblocked now that #7
 - [x] Architect URL construction fixed (#717 — closed)
 - [x] Architect pitch output fixed (#716 — closed)
 - [x] Architect false-completion bug fixed (#735 — closed)
-- [ ] close_vision_issue error handling (#737 — backlog, priority, tech-debt)
+- [x] close_vision_issue error handling (#737 — closed)
-Status: IN PROGRESS — #737 is a follow-on reliability fix, architect otherwise functional
+Status: DONE
 
 ## Objective: Adoption — Dev agent pipeline reliability
 - [x] Dev-poll deadlock in same-container case (#693 — closed)
 - [x] Profile journal digestion hang (#702 — closed)
-Status: DONE
+- [ ] Dev-poll post-crash deadlock — self-assigned issue never recovered (#749 — in-progress, priority)
+Status: IN PROGRESS — #749 being worked by dev-bot
 
 ## Objective: Adoption — Assistant (disinto-chat) (#623)
 - [x] Caddy subpath routing skeleton (#704 — closed)
@@ -100,9 +108,10 @@ Status: DONE
 - [x] Cost caps + rate limiting (#711 — closed)
 - [x] Conversation history persistence (#710 — closed)
 - [x] Per-project subdomain fallback (#713 — closed)
-- [ ] Claude identity isolation (#707 — backlog, relabeled for ci_exhausted retry)
+- [ ] Claude identity isolation (#707 — blocked, pending #742 CI investigation)
-- [ ] Escalation tools (#712 — backlog, priority, relabeled for ci_exhausted retry, has PR #733)
+- [ ] Escalation tools (#712 — blocked, priority, pending #742 CI investigation)
-Status: IN PROGRESS — 8/10 sub-issues closed, 2 relabeled for retry
+- [ ] CI exhaustion investigation (#742 — in-progress, priority, assigned to dev-qwen)
+Status: IN PROGRESS — 8/10 sub-issues closed, #742 investigating CI failures blocking #707/#712
 
 ## Objective: Ship (Fold 2) — Deploy profiles per artifact type
 - [ ] No deploy profiles defined

vault/pending/disinto-caddy-ssh-key.md

@@ -0,0 +1,29 @@
+# Request: Caddy host SSH key for engagement data collection
+
+## What
+Provision a dedicated SSH keypair with `command=` restriction on the Caddy host serving disinto.ai. This key allows the edge container to fetch access logs for engagement measurement.
+
+## Why
+Blocks #745 (collect-engagement formula) and #426 (website observability).
+The factory cannot make disinto.ai an observable without reading access logs from the Caddy host.
+Waiting since 2026-04-14.
+
+## Human action
+1. Generate keypair: `ssh-keygen -t ed25519 -f caddy-collect -N '' -C 'disinto-collect-engagement'`
+2. Install public key on Caddy host in `~/.ssh/authorized_keys` with restriction:
+   ```
+   command="cat /var/log/caddy/access.log",no-port-forwarding,no-X11-forwarding,no-agent-forwarding ssh-ed25519 AAAA... disinto-collect-engagement
+   ```
+3. Ensure Caddy is configured for JSON structured access log format
+4. Add private key as `CADDY_SSH_KEY` to `.env.vault.enc`
+5. Add Caddy host address as `CADDY_HOST` to `.env.vault.enc`
+6. Test: `ssh -i caddy-collect user@caddy-host` should output the access log and disconnect
+
+## Factory will then
+- Run `collect-engagement` formula daily via edge container cron
+- Commit evidence JSON to `evidence/engagement/YYYY-MM-DD.json` in ops repo
+- Planner uses evidence data to assess landing page effectiveness
+
+## Unblocks
+- #745 — collect-engagement formula + container script
+- #426 — website observability (parent vision issue)