@ghostwater/soulforge 0.7.0 → 0.8.0

package/workflows/feature-dev/workflow.yml

@@ -9,15 +9,15 @@ description: |
   verifier checks → integration testing → PR → final review.
 
 defaults:
-  executor: claude-code
-  model: opus
+  executor: "{{executor:codex-cli}}"
+  model: "{{model:gpt-5.3-codex}}"
   timeout: 600
   max_retries: 2
 
 steps:
   - id: plan
-    executor: claude-code
-    model: opus
+    executor: "{{executor:codex-cli}}"
+    model: "{{model:gpt-5.3-codex}}"
     workdir: "{{workdir}}"
     output_schema:
       fields:
@@ -59,17 +59,17 @@ steps:
 
       Instructions:
       1. Explore the codebase to understand the stack, conventions, and patterns
-      2. Break the task into small user stories (max 20)
+      2. Break the task into the minimum number of small user stories needed.
+         1-3 stories is perfectly fine for simple tasks, and a single story is explicitly allowed when the task is simple.
+         Most tasks should be on the lower end (typically 1-5 stories).
+         20 stories is an absolute hard ceiling, not a target.
+         Do NOT pad with documentation, logging, config, or cleanup stories unless explicitly requested.
       3. Order by dependency: schema/DB first, backend, frontend, integration
       4. Each story must fit in one developer session (one context window)
       5. Every acceptance criterion must be mechanically verifiable
       6. Always include "Typecheck passes" as the last criterion in every story
       7. Every story MUST include test criteria
 
-      Reply with:
-      STATUS: done
-      STORIES_JSON: [ ... array of story objects ... ]
-    expects: "STATUS: done"
 
   - id: review-plan
     executor: self
@@ -90,13 +90,13 @@ steps:
       STORIES: {{stories_json}}
 
   - id: implement
-    executor: claude-code
-    model: opus
+    executor: "{{executor:codex-cli}}"
+    model: "{{model:gpt-5.3-codex}}"
     workdir: "{{workdir}}"
     notify: [on_complete, on_fail]
     type: loop
     loop:
-      over: stories
+      over: plan.stories
       completion: all_done
       fresh_session: true
       verify_each: true
@@ -120,8 +120,7 @@ steps:
 
       TASK (overall): {{task}}
       WORKDIR: {{workdir}}
-      BUILD_CMD: {{build_cmd}}
-      TEST_CMD: {{test_cmd}}
+      Build/Test commands: discover from AGENTS.md and repo scripts.
 
       CURRENT STORY:
       {{current_story}}
@@ -142,15 +141,10 @@ steps:
       5. Run tests
       6. Commit: feat: {{current_story_id}} - {{current_story_title}}
 
-      Reply with:
-      STATUS: done
-      CHANGES: what you implemented
-      TESTS: what tests you wrote
-    expects: "STATUS: done"
 
   - id: verify
-    executor: claude-code
-    model: opus
+    executor: "{{executor:codex-cli}}"
+    model: "{{model:gpt-5.3-codex}}"
     workdir: "{{workdir}}"
     output_schema:
       fields:
@@ -175,7 +169,7 @@ steps:
 
       WORKDIR: {{workdir}}
       CHANGES: {{changes}}
-      TEST_CMD: {{test_cmd}}
+      Build/Test commands: discover from AGENTS.md and repo scripts.
 
       CURRENT STORY:
       {{current_story}}
@@ -184,22 +178,13 @@ steps:
       1. Code exists (not just TODOs or placeholders)
       2. Each acceptance criterion is met
       3. Tests were written
-      4. Tests pass (run {{test_cmd}})
-      5. Typecheck passes
+      4. Tests pass (using the project's standard test command)
+      5. Typecheck/build passes (using the project's standard build/typecheck command)
 
-      Reply with:
-      STATUS: done
-      VERIFIED: What you confirmed
-
-      Or if incomplete:
-      STATUS: retry
-      ISSUES:
-      - What's missing
-    expects: "STATUS: done"
 
   - id: test
-    executor: claude-code
-    model: opus
+    executor: "{{executor:codex-cli}}"
+    model: "{{model:gpt-5.3-codex}}"
     workdir: "{{workdir}}"
     output_schema:
       fields:
@@ -216,20 +201,16 @@ steps:
 
       TASK: {{task}}
       WORKDIR: {{workdir}}
-      TEST_CMD: {{test_cmd}}
+      Build/Test commands: discover from AGENTS.md and repo scripts.
 
       1. Run the full test suite
       2. Look for integration issues between stories
       3. Check error handling and edge cases
 
-      Reply with:
-      STATUS: done
-      RESULTS: What you tested and outcomes
-    expects: "STATUS: done"
 
   - id: pr
-    executor: claude-code
-    model: opus
+    executor: "{{executor:codex-cli}}"
+    model: "{{model:gpt-5.3-codex}}"
     workdir: "{{workdir}}"
     notify: on_complete
     output_schema:
@@ -255,16 +236,11 @@ steps:
 
       Create a PR with gh pr create. Capture the PR number.
 
-      Reply with:
-      STATUS: done
-      PR: URL to the pull request
-      PR_NUMBER: <number only, e.g. 42>
-    expects: "STATUS: done"
 
   # ── Automated Review Loop ──────────────────────────────────────────
   - id: code-review
-    executor: claude-code
-    model: opus
+    executor: "{{executor:codex-cli}}"
+    model: "{{model:gpt-5.3-codex}}"
     workdir: "{{workdir}}"
     output_schema:
       fields:
@@ -285,15 +261,14 @@ steps:
 
       WORKDIR: {{workdir}}
       TASK: {{task}}
-      BUILD_CMD: {{build_cmd}}
-      TEST_CMD: {{test_cmd}}
+      Build/Test commands: discover from AGENTS.md and repo scripts.
 
       Instructions:
       1. Read ALL changed files in the PR
       2. Check for correctness, edge cases, potential bugs
       3. Check test quality and coverage
-      4. Run build: {{build_cmd}}
-      5. Run tests: {{test_cmd}}
+      4. Run the project's standard build/typecheck command
+      5. Run the project's standard test command
       6. Post ALL findings as a single comment on the PR:
            gh pr comment {{pr_number}} --body "<your review>"
       7. Flag severity for each finding (High/Medium/Low)
@@ -303,7 +278,6 @@ steps:
       Reply with exactly one of:
         REVIEW_DECISION: pass
         REVIEW_DECISION: fix
-    expects: "REVIEW_DECISION:"
 
   - id: review-gate
     type: gate
@@ -345,8 +319,8 @@ steps:
       max_loops: 5
 
   - id: review-fix
-    executor: claude-code
-    model: opus
+    executor: "{{executor:codex-cli}}"
+    model: "{{model:gpt-5.3-codex}}"
     workdir: "{{workdir}}"
     output_schema:
       fields:
@@ -368,14 +342,13 @@ steps:
         gh pr view {{pr_number}} --comments --json comments --jq '.comments[-1].body'
 
       WORKDIR: {{workdir}}
-      BUILD_CMD: {{build_cmd}}
-      TEST_CMD: {{test_cmd}}
+      Build/Test commands: discover from AGENTS.md and repo scripts.
 
       Fix ONLY the items marked for fixing. Do not touch anything else.
 
       After fixing:
-      1. Run build: {{build_cmd}}
-      2. Run tests: {{test_cmd}}
+      1. Run the project's standard build/typecheck command
+      2. Run the project's standard test command
       3. Commit and push
 
       DO NOT:
@@ -383,10 +356,6 @@ steps:
       - Re-implement the original feature
       - Touch files not related to the review findings
 
-      Reply with:
-      STATUS: done
-      CHANGES: what you fixed
-    expects: "STATUS: done"
     next: code-review
 
   - id: final-review