fix(decisions-log): enumerate decisions.yaml writes in every phase agent (closes #399)

.claude-plugin/marketplace.json

@@ -6,13 +6,13 @@
     "url": "https://github.com/jjackson"
   },
   "metadata": {
-    "version": "0.13.326"
+    "version": "0.13.327"
   },
   "plugins": [
     {
       "name": "ace",
       "source": "./",
-      "version": "0.13.326",
+      "version": "0.13.327",
       "description": "AI Connect Engine — orchestrates the CRISPR-Connect lifecycle from idea through app building, Connect setup, LLO management, and closeout"
     }
   ]

.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "ace",
-  "version": "0.13.326",
+  "version": "0.13.327",
   "description": "AI Connect Engine — orchestrates the CRISPR-Connect lifecycle from idea through app building, Connect setup, LLO management, and closeout",
   "author": {
     "name": "Jonathan Jackson",

VERSION

@@ -1 +1 @@
-0.13.326
+0.13.327

agents/closeout.md

@@ -24,7 +24,9 @@ You handle the closeout of a completed CRISPR-Connect opportunity.
 ### Step 1: Invoice and Payment
 Invoke the `opp-closeout` skill.
 - Input: opportunity details, invoice data
-- Output: invoices pulled, Jira payment ticket created
+- Output:
+  - invoices pulled, Jira payment ticket created
+  - Appended `closeout-depth`, `learnings-summary-scope` rows in `decisions.yaml` (merge-only; bar criterion per `skills/idea-to-pdd/SKILL.md § Decisions Log Convention`).
 
 ### Step 2: LLO Feedback
 Invoke the `llo-feedback` skill.

agents/commcare-setup.md

@@ -178,7 +178,9 @@ Step 1.5), since `nova-plugin#2` affects every architect dispatch
 identically — not just builds.
 
 - Input: approved PDD from GDrive
-- Output: app JSON/CCZ files + summaries written to `ACE/<opp-name>/app-summaries/`
+- Output:
+  - app JSON/CCZ files + summaries written to `ACE/<opp-name>/app-summaries/`
+  - From `pdd-to-deliver-app`: appended `deliver-unit-count`, `one-form-per-module-workaround`, `multimedia-coverage-strategy` rows in `decisions.yaml` (merge-only; rows are emitted only when they meet the bar criterion in `skills/idea-to-pdd/SKILL.md § Decisions Log Convention` — the list is a working catalog, not a required set).
 - **LLM-as-Judge:** unless `--no-evals` was passed, dispatch
   `pdd-to-learn-app-eval` after the Learn build and
   `pdd-to-deliver-app-eval` after the Deliver build. Each writes
@@ -277,7 +279,10 @@ pre-flight (real failure mode from turmeric run 20260509-0455). See
 `agents/ace-orchestrator.md § Skill Invocation Discipline`.
 
 - Reads: pdd-to-app-journeys.md, both app summaries, Nova blueprints
-- Writes: app-test-cases.yaml + recipes/J*.yaml under app-test-cases/
+- Writes:
+  - app-test-cases.yaml + recipes/J*.yaml under app-test-cases/
+  - Appended `test-scenario-count`, `test-archetype-coverage` rows in `decisions.yaml` (merge-only; bar criterion per `skills/idea-to-pdd/SKILL.md § Decisions Log Convention`).
+    Note: `app-test-cases/SKILL.md` records these rows with `phase: 6-qa-and-training` even though dispatch happens here in Phase 3. That's an upstream `phase:` mismatch in the SKILL.md (see issue #399 follow-up); don't paper over it on the writer side.
 - Halts on missing inputs or recipe-validation failure
 
 Phase 6 shallow runs the smoke recipes; /ace:qa-deep runs them all.

agents/connect-setup.md

@@ -54,11 +54,13 @@ Invoke the `connect-program-setup` skill.
 Invoke the `connect-opp-setup` skill.
 
 - **Input:** program UUID from Step 1; PDD; deployment summary from Phase 3.
-- **Output:** Opportunity created with `is_test=true`, verification flags +
-  payment units configured, **activated**, and ACE test user
-  (`${ACE_E2E_PHONE}`) pre-invited. Details in
-  `ACE/<opp-name>/runs/<run-id>/4-connect/connect-opp-setup.md` with the
-  opportunity UUID.
+- **Output:**
+  - Opportunity created with `is_test=true`, verification flags +
+    payment units configured, **activated**, and ACE test user
+    (`${ACE_E2E_PHONE}`) pre-invited. Details in
+    `ACE/<opp-name>/runs/<run-id>/4-connect/connect-opp-setup.md` with the
+    opportunity UUID.
+  - Appended `verification-flags`, `payment-unit-shape`, `opportunity-end-date` rows in `decisions.yaml` (merge-only; bar criterion per `skills/idea-to-pdd/SKILL.md § Decisions Log Convention` — only rows that meet the bar are emitted).
 - **Depends on:** Step 1 (needs program UUID); Phase 3 outputs (needs
   CommCare app metadata).
 - **Activation:** Phase 4 activates the opp synchronously (Step 6.5 in

agents/execution-manager.md

@@ -71,7 +71,9 @@ Invoke the `llo-uat` skill.
 ### Step 3: Opportunity Go-Live
 Invoke the `llo-launch` skill.
 - Input: UAT results confirming LLO sign-offs
-- Output: opportunity activated in Connect, LLOs notified of go-live
+- Output:
+  - opportunity activated in Connect, LLOs notified of go-live
+  - Appended `llo-capacity-actual`, `day-one-readiness`, `downstream-handoff-alignment`, `stop-loss-planning` rows in `decisions.yaml` (merge-only; bar criterion per `skills/idea-to-pdd/SKILL.md § Decisions Log Convention`). These four rows map 1:1 to `llo-launch-eval`'s viability axis (PR #145) so the rubric has structured input instead of grading on prose.
 - **Gate (review mode):** Present launch readiness summary for approval before activating
 - **LLM-as-Judge:** unless `--no-evals` was passed, dispatch
   `llo-launch-eval` after activation. Writes `verdicts/llo-launch.yaml`.

agents/ocs-setup.md

@@ -36,9 +36,11 @@ refuses to proceed without a fresh, passing deep verdict.
 ### Step 1: Configure the chatbot
 Invoke the `ocs-agent-setup` skill.
 - Input: `ACE/<opp-name>/` — PDD, training materials, app summaries, opportunity config
-- Output: cloned chatbot with opp system prompt, RAG collection indexed,
-  version published. `ACE/<opp-name>/runs/<run-id>/5-ocs/ocs-agent-setup.md` written with
-  `{experiment_id, public_id, embed_key, collection_id, pipeline_id, version_number}`
+- Output:
+  - cloned chatbot with opp system prompt, RAG collection indexed,
+    version published. `ACE/<opp-name>/runs/<run-id>/5-ocs/ocs-agent-setup.md` written with
+    `{experiment_id, public_id, embed_key, collection_id, pipeline_id, version_number}`
+  - Appended `system-prompt-baseline`, `rag-collection-scope`, `test-prompt-count` rows in `decisions.yaml` (merge-only; bar criterion per `skills/idea-to-pdd/SKILL.md § Decisions Log Convention`).
 - Idempotent: if a bot named `"ACE - <opp-name>"` already exists, resumes from
   existing config
 

agents/orchestrator-reference.md

@@ -603,9 +603,26 @@ Scope and `skills/idea-to-pdd/SKILL.md` § Decisions Log Convention §
 Bar criterion). Each phase's primary writing skill owns the rows it
 writes. The orchestrator stub-fills + warns post-phase if a phase
 wrote zero rows AND the calibration set for that phase has any
-required rows. PR #1 covers Phase 1 (`idea-to-pdd`); Phase 3–10 writes
-ship in PR #3 of the decisions-log series. Schema and YAML helpers
-live in `lib/decisions-schema.ts`.
+required rows. Schema and YAML helpers live in `lib/decisions-schema.ts`.
+
+**The procedural authority for each phase is the per-step `Output`
+block in its `agents/<phase>.md` file**, not the catalog in the writing
+skill's `SKILL.md`. The catalog (the `## Decisions Log` section in each
+producer skill) is a teaching device — listing the rows that commonly
+qualify under the bar for that phase. The agent file's per-step `Output`
+bullets are what the dispatched subagent treats as its checklist.
+Documented catalogs without a matching per-step bullet produced silent
+zero-write failures across Phase 2–9 on the malaria-itn-app run filed
+as `jjackson/ace#399`; the fix was the per-step Output enumeration that
+now lives in every downstream agent file. When you add a new skill that
+writes anchor rows, BOTH must be updated together: the catalog in
+`SKILL.md` AND the `Output:` bullet in the dispatching agent file. The
+catalog alone is not load-bearing.
+
+History: PR #1 of the decisions-log series shipped Phase 1
+(`idea-to-pdd` + `pdd-to-work-order`). PR #4 added Decisions Log
+catalogs to Phase 3–10 skills but did NOT update the agent files; that
+gap was closed by the issue #399 fix.
 
 ## Recurring writers — TBD
 

agents/solicitation-management.md

@@ -41,14 +41,16 @@ lifecycle requires explicit human approval and is run manually via
 Invoke the `solicitation-create` skill.
 - Input: approved PDD (`inputs/pdd.md`), `opp.yaml`
   (`connect.program.id`, total_budget)
-- Output: `solicitation/published.md`,
-  `phases.solicitation-management.products.solicitation` populated in
-  the current run's `run_state.yaml` with
-  `{solicitation_id, public_url, deadline, status: open,
-  labs_program_id, ...}`. `selected_llo` is populated by
-  `solicitation-review` on award at `products.selected_llo`. The labs
-  program int is cached durably at
-  `opp.yaml.connect.program.labs_int_id`.
+- Output:
+  - `solicitation/published.md`,
+    `phases.solicitation-management.products.solicitation` populated in
+    the current run's `run_state.yaml` with
+    `{solicitation_id, public_url, deadline, status: open,
+    labs_program_id, ...}`. `selected_llo` is populated by
+    `solicitation-review` on award at `products.selected_llo`. The labs
+    program int is cached durably at
+    `opp.yaml.connect.program.labs_int_id`.
+  - Appended `solicitation-type`, `response-deadline`, `response-template-choice` rows in `decisions.yaml` (merge-only; bar criterion per `skills/idea-to-pdd/SKILL.md § Decisions Log Convention`).
 - **LLM-as-Judge:** unless `--no-evals` was passed, dispatch
   `solicitation-create-eval` after publish. Writes
   `verdicts/solicitation-create.yaml`.

agents/synthetic-data-and-workflows.md

@@ -113,6 +113,7 @@ Invoke `synthetic-narrative-plan`.
   opp.yaml.
 - Produces `7-synthetic/synthetic-narrative-plan.md` (human narrative)
   and `7-synthetic/synthetic-narrative-plan.yaml` (the manifest).
+- Appends `persona-count`, `scenario-count`, `narrative-arc-shape` rows in `decisions.yaml` (merge-only; bar criterion per `skills/idea-to-pdd/SKILL.md § Decisions Log Convention`).
 - The manifest schema is identical to `synthetic-data-generate`'s; this
   skill just authors a richer instance with named FLWs, deliberate
   anomalies, coaching-arc transcripts.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "ace",
-  "version": "0.13.326",
+  "version": "0.13.327",
   "description": "AI Connect Engine - orchestrator for building Connect Opps using AI",
   "type": "module",
   "scripts": {