feat(bmm): draft bmad-dev-auto skill

Introduce a separate machine-first implementation workflow for bmad-auto sessions without layering more automation branches into bmad-quick-dev.

src/bmm-skills/4-implementation/bmad-dev-auto/SKILL.md

@@ -0,0 +1,105 @@
+---
+name: bmad-dev-auto
+description: 'Implements one story or deferred-work bundle unattended for a bmad-auto orchestrator. Use when a machine-run session must execute development work with no human interaction.'
+---
+
+# BMad Dev Auto
+
+**Goal:** turn one orchestrator task into verified code plus on-disk artifacts the orchestrator can inspect.
+
+This skill is for unattended runs only. It is not a variant of `bmad-quick-dev`; it is a separate machine-first workflow.
+
+## Contract
+
+- No greeting.
+- No questions.
+- No menus.
+- No editor.
+- No commit or push.
+- `result.json` is the last successful action.
+- If blocked, write `escalation.json`, write `result.json`, and end the turn.
+
+## Invocation
+
+The orchestrator invokes one of:
+
+- `<story-key>`
+- `<story-key> --feedback <path>`
+- `--dw-bundle <path>`
+- `--dw-bundle <path> --feedback <path>`
+
+Environment:
+
+- `BMAD_AUTO_RUN_DIR`
+- `BMAD_AUTO_TASK_ID`
+- optional `BMAD_AUTO_SKIP_REVIEW=1`
+
+Files:
+
+- result file: `$BMAD_AUTO_RUN_DIR/tasks/$BMAD_AUTO_TASK_ID/result.json`
+- escalation file: `$BMAD_AUTO_RUN_DIR/tasks/$BMAD_AUTO_TASK_ID/escalation.json`
+
+## Conventions
+
+- Bare paths resolve from the skill root.
+- `{skill-root}` is this installed skill directory.
+- `{project-root}` is the working directory.
+- `{skill-name}` is the skill directory basename.
+- `{workflow.<name>}` comes from the merged `customize.toml` `[workflow]` table.
+
+## On Activation
+
+1. Resolve customization:
+   `python3 {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+2. If that fails, merge these in order using BMad structural merge rules:
+   - `{skill-root}/customize.toml`
+   - `{project-root}/_bmad/custom/{skill-name}.toml`
+   - `{project-root}/_bmad/custom/{skill-name}.user.toml`
+3. Run each `{workflow.activation_steps_prepend}` entry.
+4. Treat each `{workflow.persistent_facts}` entry as persistent context for the whole run.
+5. Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+   - `project_name`
+   - `planning_artifacts`
+   - `implementation_artifacts`
+   - `communication_language`
+   - `document_output_language`
+   - `user_skill_level`
+   - `date`
+   - `sprint_status` = `{implementation_artifacts}/sprint-status.yaml`
+6. Run each `{workflow.activation_steps_append}` entry.
+
+## Rules
+
+- Speak tersely. Spend tokens on work, not narration.
+- Never wait for user input.
+- Treat the invocation as authoritative input.
+- Spec target is **1500-4000 tokens**. On real multi-goal scope, split and defer the rest.
+- Preserve anything inside `<frozen-after-approval>` once the spec is approved.
+- Use the full `git rev-parse HEAD` hash for `baseline_commit`.
+
+## Result Schema
+
+```json
+{
+  "workflow": "dev-auto",
+  "story_key": "<story key or null>",
+  "spec_file": "<absolute path>",
+  "baseline_commit": "<full hash or NO_VCS>",
+  "status": "in-review|done|blocked",
+  "tasks_total": 0,
+  "tasks_done": 0,
+  "verification": [{"command": "<cmd>", "ok": true}],
+  "escalations": [{"type": "<kind>", "severity": "CRITICAL|PREFERENCE", "detail": "<detail>"}],
+  "dw_ids": ["DW-1"]
+}
+```
+
+`status` means:
+
+- `in-review`: code complete; a separate review run is expected.
+- `done`: no separate review run is expected.
+- `blocked`: the task could not continue safely.
+
+## First Step
+
+Read fully and follow `./step-01-resolve.md`.

src/bmm-skills/4-implementation/bmad-dev-auto/customize.toml

@@ -0,0 +1,12 @@
+# DO NOT EDIT -- overwritten on every update.
+
+[workflow]
+
+activation_steps_prepend = []
+activation_steps_append = []
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+on_complete = ""

src/bmm-skills/4-implementation/bmad-dev-auto/deferred-work-format.md

@@ -0,0 +1,35 @@
+# Deferred Work Format
+
+Append-only format for `{implementation_artifacts}/deferred-work.md`.
+
+## Rules
+
+- Do not rewrite or delete old entries.
+- Before appending, scan for an equivalent open item.
+- If one exists, add `seen-again:` instead of duplicating it.
+- Number entries `DW-1`, `DW-2`, and so on.
+
+## Entry
+
+```markdown
+### DW-<seq>: <title>
+
+origin: <workflow + artifact + date>
+location: <file:line | component | n/a>
+severity: <critical | high | medium | low>
+reason: <why it is deferred>
+status: open
+```
+
+When completed, change `status:` to `done <date>` and add:
+
+```markdown
+resolution: <one line>
+```
+
+Optional:
+
+```markdown
+seen-again: <date and source>
+decision: <date> <label> -- <detail>
+```

src/bmm-skills/4-implementation/bmad-dev-auto/spec-template.md

@@ -0,0 +1,51 @@
+---
+title: '{title}'
+type: 'feature'
+created: '{date}'
+status: 'draft'
+context: []
+baseline_commit: ''
+---
+
+<!-- Target: 1500-4000 tokens in auto mode. Remove comments when filling. -->
+
+<frozen-after-approval reason="orchestrator-owned intent — do not modify after approval unless a non-frozen spec defect must be corrected">
+
+## Intent
+
+**Problem:** ONE_TO_TWO_SENTENCES
+
+**Approach:** ONE_TO_TWO_SENTENCES
+
+## Boundaries & Constraints
+
+**Always:** INVARIANT_RULES
+
+**Never:** NON_GOALS_AND_FORBIDDEN_APPROACHES
+
+## I/O & Edge Cases
+
+| Scenario | Input / State | Expected Behavior | Error Handling |
+|----------|----------------|-------------------|----------------|
+| HAPPY_PATH | INPUT | OUTCOME | N/A |
+
+</frozen-after-approval>
+
+## Code Map
+
+- `FILE` -- ROLE
+
+## Tasks & Acceptance
+
+**Execution:**
+- [ ] `FILE` -- ACTION -- RATIONALE
+
+**Acceptance Criteria:**
+- Given PRECONDITION, when ACTION, then EXPECTED_RESULT
+
+## Spec Change Log
+
+## Verification
+
+**Commands:**
+- `COMMAND` -- expected: SUCCESS

src/bmm-skills/4-implementation/bmad-dev-auto/step-01-resolve.md

@@ -0,0 +1,37 @@
+---
+---
+
+# Step 1: Resolve Task
+
+## Instructions
+
+1. Parse the invocation into one of:
+   - story mode
+   - story + feedback mode
+   - bundle mode
+   - bundle + feedback mode
+2. Set:
+   - `{result_file}` = `$BMAD_AUTO_RUN_DIR/tasks/$BMAD_AUTO_TASK_ID/result.json`
+   - `{escalation_file}` = `$BMAD_AUTO_RUN_DIR/tasks/$BMAD_AUTO_TASK_ID/escalation.json`
+3. If required environment variables are missing, write a `CRITICAL` escalation (`type: missing-env`) and end the run.
+4. If in bundle mode:
+   - read the bundle file first
+   - set `{bundle_name}`
+   - set `{dw_ids}` from the bundle file
+   - set `{story_key}` = `dw-{bundle_name}`
+   - set `{spec_file}` = `{implementation_artifacts}/spec-dw-{bundle_name}.md`
+5. Otherwise:
+   - set `{story_key}` from the invocation
+   - derive `{epic_num}` and `{story_num}` from its leading numeric segments
+   - set `{spec_file}` = `{implementation_artifacts}/spec-{story_key}.md`
+6. If a feedback file was passed, read it before deciding the route.
+7. Route:
+   - if feedback mode and `{spec_file}` exists: go to step 3
+   - else if `{spec_file}` exists with `status: draft`: go to step 2
+   - else if `{spec_file}` exists with `status: ready-for-dev|in-progress|in-review|done`: go to step 3
+   - else: go to step 2
+
+## Next
+
+- Step 2: `./step-02-plan.md`
+- Step 3: `./step-03-implement.md`

src/bmm-skills/4-implementation/bmad-dev-auto/step-02-plan.md

@@ -0,0 +1,28 @@
+---
+deferred_work_file: '{implementation_artifacts}/deferred-work.md'
+---
+
+# Step 2: Plan
+
+## Instructions
+
+1. Investigate the codebase and relevant context files.
+2. Read `./spec-template.md` fully.
+3. Write `{spec_file}` from the template.
+4. Self-review the spec:
+   - one goal
+   - actionable tasks
+   - testable acceptance criteria
+   - no placeholders
+5. If intent is still unclear, write a `CRITICAL` escalation (`type: intent-gap`) and end the run.
+6. If the scope is truly multi-goal or the spec exceeds 4000 tokens:
+   - keep the main goal in `{spec_file}`
+   - append each deferred secondary goal to `{deferred_work_file}` using `./deferred-work-format.md`
+   - rewrite the spec to match the narrowed scope
+7. Re-read `{spec_file}` from disk.
+8. If it is missing or empty, write a `CRITICAL` escalation (`type: spec-write-failure`) and end the run.
+9. Set the spec frontmatter `status:` to `ready-for-dev`.
+
+## Next
+
+Read fully and follow `./step-03-implement.md`.

src/bmm-skills/4-implementation/bmad-dev-auto/step-03-implement.md

@@ -0,0 +1,21 @@
+---
+---
+
+# Step 3: Implement
+
+## Instructions
+
+1. Verify `{spec_file}` exists and is non-empty. If not, write a `CRITICAL` escalation (`type: missing-spec`) and end the run.
+2. Capture `baseline_commit` into the spec frontmatter:
+   - full `git rev-parse HEAD`
+   - or `NO_VCS` if git is unavailable
+3. If this is not repair mode against an already-`done` spec, set the spec `status:` to `in-progress`.
+4. If this is not bundle mode, follow `./sync-sprint-status.md` with `{target_status}` = `in-progress`.
+5. Load any spec `context:` files.
+6. Implement the spec directly or via sub-agents.
+7. Mark every completed task in `## Tasks & Acceptance` as `[x]`.
+8. If any listed task remains incomplete, finish it before continuing.
+
+## Next
+
+Read fully and follow `./step-04-finalize.md`.

src/bmm-skills/4-implementation/bmad-dev-auto/step-04-finalize.md

@@ -0,0 +1,22 @@
+---
+deferred_work_file: '{implementation_artifacts}/deferred-work.md'
+---
+
+# Step 4: Finalize
+
+## Instructions
+
+1. Verify every task in `## Tasks & Acceptance` is checked.
+2. Run every command in the spec `## Verification` section.
+3. If a command fails and you cannot fix it without violating the frozen intent, write a `CRITICAL` escalation (`type: verification-failure`) and end the run.
+4. Set final spec status:
+   - `done` when `BMAD_AUTO_SKIP_REVIEW=1`
+   - `in-review` otherwise
+5. If not bundle mode, follow `./sync-sprint-status.md` with:
+   - `{target_status}` = `done` when `BMAD_AUTO_SKIP_REVIEW=1`
+   - `{target_status}` = `review` otherwise
+6. If bundle mode, update every referenced deferred-work entry:
+   - `status: done <date>`
+   - `resolution: <one line>`
+7. Write `{result_file}` using the schema from `SKILL.md`.
+8. End the turn with a one-line outcome.

src/bmm-skills/4-implementation/bmad-dev-auto/sync-sprint-status.md

@@ -0,0 +1,18 @@
+# Sync Sprint Status
+
+Skip this file if:
+
+- `{story_key}` is unset
+- `{story_key}` starts with `dw-`
+- `{sprint_status}` does not exist
+
+## Instructions
+
+1. Load the full `{sprint_status}` file.
+2. Find `development_status[{story_key}]`.
+3. If missing, return without writing.
+4. Never regress status.
+5. Set the story to `{target_status}`.
+6. If `{target_status}` is `in-progress`, derive `epic-{N}` from the story key and lift it from `backlog` to `in-progress` when present.
+7. Refresh `last_updated`.
+8. Save while preserving comments and structure.