chore: planner run 2026-04-17

- Updated prerequisite tree: #425 closed, added Nomad backend and
  hire-an-agent objectives, tracked S2 verification bugs
- Filed vault procurement for #758 (ops repo branch protection)
- Memory update (runs 3-13 summarized): constraint focus shifted to
  #758 root blocker and hire-an-agent cascading bugs

knowledge/planner-memory.md

@@ -1,28 +1,37 @@
-<!-- summarized-through-run: 2 -->
+<!-- summarized-through-run: 13 -->
 # Planner Memory
-Last updated: 2026-04-08
+Last updated: 2026-04-17
 
 ## Run history
 - Run 1 (2026-04-05): First planning run. Established prerequisite tree from VISION.md.
-- Run 2 (2026-04-08): Foundation milestone completed. All 5 blockers resolved. Triaged 4 predictor predictions (all actioned). Filed #425 (ops repo migration) and #426 (website observability). Focus shifted to Adoption milestone.
+- Run 2 (2026-04-08): Foundation milestone completed. Triaged 4 predictions. Filed #425, #426.
+- Runs 3-12 (2026-04-12 to 2026-04-15): Maintained tree locally. Ops repo push blocked by #758 (branch protection). Nomad migration sprint filed and tracked. hire-an-agent bugs surfaced.
+- Run 13 (2026-04-17): Updated tree with Nomad migration and hire-an-agent objectives. Filed vault procurement for #758. Memory update (10 runs since last summary).
 
 ## Current constraint focus
-1. **Ops repo migration** (#425) — existing deployments missing evidence/, sprints/, portfolio.md, RESOURCES.md, vault subdirs. Blocks evidence collection for all agents. Priority+backlog.
+1. **Ops repo branch protection** (#758) — HUMAN_BLOCKED. All agent writes to ops repo frozen since 2026-04-08. Vault procurement filed (vault/pending/disinto-ops-branch-protection.md). Gardener flagged 2x. Root blocker for Adoption.
-2. **Website observability** (#426) — disinto.ai has no analytics/measurement. Blocks Adoption landing page and Ship engagement measurement. Vision-level (design decisions needed).
+2. **hire-an-agent defects** (#834 umbrella) — compose generator, TOML parser, collaborator setup, image rebuild (#887). Blocks multi-agent parallel operation. Multiple priority sub-issues in-progress.
-3. **Example project** — none exists. Blocked on bootstrap verification after #425.
+3. **issue_claim race** (#830) — verify assignee after PATCH to prevent duplicate work. Priority, in-progress.
-4. **Deploy profiles** — not started. Vision-level design needed for per-artifact-type deployment.
+4. **Re-seed ops dirs** (#820) — blocked on #758. Once branch protection resolved, seeds vault/evidence/sprint dirs.
-5. **Assumptions register** — not started. Ship (Fold 2) prerequisite.
+5. **Website observability** (#426) — vision-level. No analytics on disinto.ai. Blocks Adoption landing page completion.
 
 ## Patterns observed
-- Foundation milestone completed in ~10 days (2026-03-28 to 2026-04-08). High bug fix velocity once dev-agent pipeline was reliable.
+- #425 closed but effect blocked by #758 — code-level fix ≠ deployment-level fix. Pattern: migration code landing without the ops repo push path being verified.
-- Predictor-planner loop working: predictor filed 4 accurate predictions, planner actioned all 4. The adversarial feedback catches stale state and planner drift effectively.
+- Nomad migration is a major new initiative (step-0 through step-2, ~20 issues). Emerged organically from architect sprints. Not in original VISION.md milestones but supports Adoption (alternative backend).
-- Predictor correctly identified planner filing Fold 2/3 vision issues (#418, #419) while Foundation was incomplete — planner must verify milestone completion before advancing.
+- hire-an-agent flow has cascading bugs: TOML parser → compose generator → entrypoint → collaborator setup. Each fix surfaces the next. Pattern: untested integration path now under first real multi-agent use.
-- Ops repo structural debt: setup_ops_repo() creates dirs for new repos but doesn't migrate existing ones. Pattern will recur with future schema changes.
+- Graph grew from 165→215 nodes, 137→259 edges. No cycles or bottlenecks. 11 orphans (benign: unused formulas/labels + orphan issue #843).
-- No observables exist — all 4 addressables (website, repo, skill, github org) lack measurement. This is the fundamental gap between Adoption and Ship.
+- Predictor has been quiet — 0 unreviewed predictions in run 13. May indicate predictor hasn't run recently or all predictions were already triaged.
+- High dev velocity on Nomad work despite ops repo write freeze — work proceeds because it's in the code repo, not ops repo.
 
 ## Strategic direction
-- Foundation DONE. Adoption is now the active milestone.
+- Foundation DONE. Adoption active with two new sub-objectives: Nomad backend and multi-agent operation.
-- #425 is the single blocking issue — once ops repo dirs exist, evidence collection unblocks and bootstrap is complete.
+- **Critical path**: #758 (human action) → #820 (re-seed) → ops repo writes resume → evidence/vault/knowledge persistence.
-- After #425: focus on example project (demonstrates full lifecycle) and website observability (#426).
+- Nomad migration is the largest active workstream but does NOT block other Adoption objectives. Track but don't gate on it.
-- Do NOT file Ship/Learn issues until Adoption objectives show progress. Predictor will check.
+- hire-an-agent bugs are the main dev-pipeline constraint — until resolved, only single-agent operation is reliable.
-- Graph report is healthy: 165 nodes, 137 edges, no cycles, no bottlenecks. 7 orphans are benign (closed issues, unused formulas/labels).
+- Do NOT file Ship/Learn issues. Adoption has 3 blocked objectives and 2 in-progress workstreams.
+- After #758: immediately push accumulated ops state, then focus on example project (#697).
+
+## Lessons learned
+- Ops repo branch protection created a 9-day knowledge blackout (2026-04-08 to 2026-04-17). Container restart would lose all intermediate state. Vault procurement items for human-blocked constraints should be filed immediately, not deferred.
+- "Code fix merged" ≠ "deployed effect realized" — verify the push path, not just the code path.
+- Umbrella issues (#834) are useful for tracking cascading bug chains in integration paths.

prerequisites.md

@@ -1,5 +1,5 @@
 # Prerequisite Tree
-<!-- Last updated: 2026-04-08 -->
+<!-- Last updated: 2026-04-17 -->
 
 ## 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 and maintaining prerequisite tree (run 1: 2026-04-05, run 2: 2026-04-08)
+- [x] Planning runs established and maintaining prerequisite tree
 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 (165 nodes, 137 edges as of 2026-04-08)
+- [x] Graph report generating successfully (215 nodes, 259 edges as of 2026-04-17)
 Status: DONE
 
 ## Objective: Foundation — Predictor-planner adversarial feedback loop
@@ -45,8 +45,11 @@ Status: DONE
 - [x] disinto init re-run stability (#158 — closed)
 - [x] disinto init repo creation API endpoint (#164 — closed)
 - [x] Prediction labels created during init (#225 — closed)
-- [ ] Ops repo migration for existing deployments (#425 — backlog+priority)
+- [x] Ops repo migration code landed (#425 — closed)
-Status: BLOCKED — #425 ops repo missing dirs on existing deployments
+- [ ] Ops repo branch protection blocks agent writes (#758 — open, HUMAN_BLOCKED, blocked-on-vault vault/pending/disinto-ops-branch-protection.md)
+- [ ] Re-seed ops repo directories on live deployment (#820 — open, blocked on #758)
+- [ ] issue_claim race — verify assignee after PATCH (#830 — open, priority)
+Status: BLOCKED — #758 ops repo branch protection requires human admin action
 
 ## Objective: Adoption — Built-in Forgejo + Woodpecker CI
 - [x] Docker compose with Forgejo + Woodpecker
@@ -54,15 +57,47 @@ Status: BLOCKED — #425 ops repo missing dirs on existing deployments
 - [x] WOODPECKER_HOST override fix (#178 — closed)
 Status: DONE
 
+## Objective: Adoption — Nomad deployment backend
+- [ ] --backend=nomad flag + stub (#821 — in-progress)
+- [ ] Nomad + systemd install (#822 — in-progress)
+- [ ] Vault install + auto-unseal (#823 — in-progress)
+- [ ] Orchestrator init (#824 — in-progress)
+- [ ] CI validation for nomad/vault artifacts (#825 — in-progress)
+- [ ] S0.1-fix: --backend swallowed as positional (#835 — in-progress)
+- [x] S0.2-fix: install.sh must install docker daemon (#871 — closed)
+- [ ] Nomad job forgejo.hcl (#840 — in-progress)
+- [ ] deploy.sh dependency-ordered job run (#841 — open)
+- [ ] Wire --with forgejo into init --backend=nomad (#842 — in-progress)
+- [ ] CI: nomad job validate (#843 — open, orphan in graph)
+- [ ] Vault policies + apply script (#879 — in-progress)
+- [ ] vault-import.sh (#880 — open)
+- [ ] Vault JWT auth + nomad workload identity (#881 — in-progress)
+- [ ] forgejo.hcl reads Vault secrets via template stanza (#882 — in-progress)
+- [ ] disinto init --import-env / --import-sops (#883 — in-progress)
+- [ ] CI: vault policy fmt + validate (#884 — backlog)
+- [ ] S2-fix bugs blocking step 2 verification (#912, #920-#922, #926, #928)
+Status: IN PROGRESS — step-0 through step-2 active, multiple S2 verification bugs
+
+## Objective: Adoption — Multi-agent parallel operation (hire-an-agent)
+- [ ] hire-an-agent + compose generator defects (#834 — in-progress, priority)
+- [ ] TOML agent section name with dash crashes load-project.sh (#862 — in-progress, priority)
+- [ ] hire-an-agent does not add collaborator (#856 — open, priority)
+- [ ] hire-an-agent TOML editor corrupts on re-run (#886 — open, priority)
+- [ ] entrypoint clone path vs TOML parse mismatch (#861 — open, priority)
+- [ ] Parametric llama bots via LLAMA_BOTS env (#831 — open, priority)
+- [ ] Generate one compose service per LLAMA_BOTS entry (#832 — backlog, priority)
+- [ ] Compose generator duplicate service name detection (#850 — blocked)
+Status: IN PROGRESS — multiple critical bugs in hire-an-agent flow
+
 ## Objective: Adoption — Landing page communicating value proposition
 - [x] Website addressable exists (disinto.ai)
 - [ ] Website observability — no engagement measurement (#426 — vision)
 Status: BLOCKED — no evidence process connected to website
 
 ## Objective: Adoption — Example project demonstrating full lifecycle
-- [ ] No example project exists
+- [ ] No example project exists (#697 — vision, priority)
-- [ ] Requires verified bootstrap (#425)
+- [ ] Requires verified bootstrap (blocked on #758 → #820)
-Status: BLOCKED — depends on bootstrap completion and ops repo migration
+Status: BLOCKED — depends on ops repo write access
 
 ## Objective: Ship (Fold 2) — Deploy profiles per artifact type
 - [ ] No deploy profiles defined
@@ -72,8 +107,8 @@ Status: BLOCKED — not started, needs design (vision-level)
 ## Objective: Ship (Fold 2) — Vault-gated fold transitions
 - [x] Vault redesign complete (#73-#77 — all closed)
 - [x] Vault PR workflow documented (docs/VAULT.md)
-- [ ] Vault directories complete in ops repo (#425 — approved/fired/rejected missing)
+- [ ] Vault directories in ops repo (#820 — blocked on #758)
-Status: BLOCKED — #425 ops repo dirs needed for vault workflow
+Status: BLOCKED — #820 ops repo dirs needed for vault workflow
 
 ## Objective: Ship (Fold 2) — Engagement measurement baked into deploy pipelines
 - [ ] No engagement measurement exists

vault/pending/disinto-ops-branch-protection.md

@@ -0,0 +1,29 @@
+# Request: Remove or relax ops repo branch protection
+
+## What
+The ops repo (`disinto-ops`) has branch protection on `main` that requires approvals, but no bot account has sufficient permissions to merge. Agents (planner, architect, supervisor) can commit locally but cannot push to remote.
+
+## Why
+Blocks #758, #820, and transitively all ops repo writes.
+- prerequisites.md has not been pushed since 2026-04-08 (run 2)
+- Vault pending items cannot be persisted
+- Sprint artifacts from architect are not pushed
+- Any container restart loses all ops repo state accumulated since 2026-04-08
+Waiting since 2026-04-15 (first gardener flag).
+
+## Human action
+1. In Forgejo, go to the `disinto-ops` repo → Settings → Branch Protection for `main`
+2. Either remove branch protection entirely (agents are the primary writers), OR add `planner-bot` to the merge whitelist
+3. Verify by pushing a test commit or confirming `planner-bot` can push directly
+
+## Factory will then
+- Push accumulated ops repo changes (prerequisites, memory, vault items, sprints)
+- Close #758 and #820
+- Re-seed missing vault directories (approved/, fired/, rejected/)
+- Resume persisting planner knowledge across container restarts
+
+## Unblocks
+- #758 — ops repo branch protection blocks all agent writes
+- #820 — re-seed ops repo directories
+- Vault workflow (pending/approved/fired/rejected lifecycle)
+- Evidence collection for all agents