@kynetic-ai/spec 0.1.2 → 0.4.0

package/templates/skills/task-work/SKILL.md

@@ -0,0 +1,296 @@
+# Task Work
+
+Structured workflow for working on tasks. Full lifecycle from start through PR merge.
+
+## When to Use
+
+- Starting work on a ready task
+- Continuing in-progress or needs_work tasks
+- Ensuring consistent task lifecycle with notes and audit trail
+
+**Not for:** Spec creation (use `/kspec:writing-specs`), plan translation (use `/kspec:plan`), or triage (use `/kspec:triage`).
+
+## Inherit Existing Work First
+
+Before starting new work, check for existing work:
+
+```bash
+kspec session start  # Shows active work at the top
+```
+
+Priority order:
+1. **needs_work** — Fix cycle: address review feedback (highest priority)
+2. **in_progress** — Continue work already started
+3. **ready (pending)** — New work to start
+
+Always inherit existing work unless explicitly told otherwise. This prevents orphaned tasks.
+
+## Task Lifecycle
+
+```
+pending → in_progress → pending_review → completed
+              ↓              ↓
+          blocked        needs_work
+                     (→ in_progress → pending_review)
+```
+
+| Command | Transition | When |
+|---------|-----------|------|
+| `kspec task start @ref` | → in_progress | Beginning work |
+| `kspec task submit @ref` | → pending_review | Code done, PR created |
+| `kspec task complete @ref --reason "..."` | → completed | PR merged |
+| `kspec task block @ref --reason "..."` | → blocked | External blocker |
+
+## Workflow Steps
+
+### 1. Choose Task
+
+```bash
+kspec tasks ready                # All ready tasks
+kspec tasks ready --eligible     # Automation-eligible only (loop mode)
+```
+
+### 2. Verify Work Is Needed
+
+Before starting, check if work is already done:
+
+```bash
+# Check git history
+git log --oneline --grep="feature-name"
+git log --oneline -- path/to/relevant/files
+
+# Check existing implementation
+kspec item get @spec-ref  # View spec and ACs
+```
+
+If already implemented: verify tests pass, AC coverage exists, then `kspec task complete @ref --reason "Already implemented"`.
+
+**Notes are context, not proof.** If a task has notes saying "already done" — verify independently. Run tests, check code, validate against ACs. If a task is in the ready queue, there's a reason.
+
+### 3. Start Task
+
+```bash
+kspec task start @ref
+```
+
+### 4. Work and Note
+
+Read all ACs (own + trait) before implementing:
+
+```bash
+kspec item get @spec-ref  # Shows own ACs AND inherited trait ACs
+```
+
+Add notes during work, not just at the end:
+
+```bash
+# Good: explains decisions and context
+kspec task note @ref "Using retry with exponential backoff. Chose 3 max retries based on API rate limits."
+
+# Bad: no context
+kspec task note @ref "Done"
+```
+
+Note when you:
+- Discover something unexpected
+- Make a design decision
+- Encounter a blocker
+- Complete a significant piece
+
+### 5. Commit
+
+Include task and spec trailers:
+
+```
+feat: add user authentication
+
+Implemented JWT-based auth with refresh tokens.
+
+Task: @task-add-auth
+Spec: @auth-feature
+```
+
+Trailers enable `kspec log @ref` to find related commits.
+
+### 6. Local Review
+
+Run quality checks before submitting. Verify:
+
+- **Own AC coverage** — Each spec AC has a test annotated `// AC: @spec-ref ac-N`
+- **Trait AC coverage** — Each inherited trait AC has a test annotated `// AC: @trait-slug ac-N`
+- **Tests pass** — Full test suite, not just new tests
+- **Code quality** — Matches existing patterns, no duplicated utilities
+- **No regressions** — Existing tests still pass
+
+```bash
+kspec validate  # Reports uncovered trait ACs as warnings
+```
+
+### 7. Submit Task
+
+```bash
+kspec task submit @ref
+```
+
+Moves task to `pending_review`. Create PR after submitting.
+
+### 8. Complete Task
+
+After PR is merged:
+
+```bash
+kspec task complete @ref --reason "Merged in PR #N. Summary of what was done."
+```
+
+## Fix Cycle
+
+When inheriting a `needs_work` task:
+
+1. **Find the PR** — Check for review comments
+   ```bash
+   gh pr list --search "Task: @task-ref" --json number,url
+   gh api repos/{owner}/{repo}/pulls/{number}/comments --jq '.[] | {path, line, body}'
+   ```
+
+2. **Fix findings** — Address MUST-FIX and SHOULD-FIX items
+
+3. **Push fixes** — Commit with descriptive message
+   ```bash
+   git add <files> && git commit -m "fix: address review feedback
+
+   Task: @task-slug"
+   git push
+   ```
+
+4. **Re-submit** — `kspec task submit @ref` (back to pending_review)
+
+You do NOT merge in a fix cycle. The reviewer handles merge decisions.
+
+## Scope Management
+
+### What's In Scope
+
+Tasks describe expected outcomes, not rigid boundaries:
+
+- **Tests need implementation?** Implementing missing functionality is in scope — the goal is verified behavior
+- **Implementation needs tests?** Proving it works is always in scope
+
+### When to Expand vs Escalate
+
+**Expand** (do it yourself):
+- Additional work is clearly implied by the goal
+- Proportional to the original task
+- You have the context
+
+**Escalate** (capture separately):
+- Scope expansion is major
+- Uncertain about the right approach
+- Outside your task's domain
+
+**When you notice something outside your task:** Capture it separately (`kspec inbox add` or `kspec task note`). Don't fix it inline — even small detours compound into drift.
+
+## AC Test Annotations
+
+Link tests to acceptance criteria:
+
+```javascript
+// AC: @spec-item ac-N
+it('should validate input', () => { ... });
+```
+
+```python
+# AC: @spec-item ac-N
+def test_validates_input():
+    ...
+```
+
+Every AC should have at least one test with this annotation.
+
+## Implementation Quality
+
+Before submitting:
+
+- **Search for existing utilities** — Don't duplicate helpers that already exist
+- **Match neighboring file style** — Naming conventions, error handling, imports
+- **Run full test suite** — Not just your new tests
+- **Validate** — `kspec validate` for spec alignment
+
+## Loop Mode
+
+Autonomous task execution without human confirmation.
+
+```bash
+kspec tasks ready --eligible  # Only automation-eligible tasks
+```
+
+### Task Selection Priority
+
+1. `needs_work` — Fix review feedback
+2. `in_progress` — Continue existing work
+3. Tasks that unblock others
+4. Highest priority ready task
+
+### Key Behaviors
+
+- Verify work is needed before starting (prevent duplicates)
+- Decisions auto-resolve without prompts
+- PR review handled externally (not this workflow)
+- All actions are logged and auditable
+
+### Blocking Rules
+
+**Block only for genuine external blockers:**
+- Human architectural decision needed
+- Spec clarification required
+- External dependency unavailable
+- Formal `depends_on` blocker
+
+**Do NOT block for:**
+- Task seems complex (do the work)
+- Tests are failing (fix them)
+- Service needs running (start it)
+
+After blocking:
+```bash
+kspec task block @ref --reason "Reason..."
+kspec tasks ready --eligible  # Check for other work
+# If tasks exist: work on the next one
+# If empty: stop responding (ralph auto-exits)
+```
+
+### Turn Completion
+
+After creating a PR, **stop responding**. Ralph continues automatically — it checks for remaining eligible tasks and exits the loop when none remain.
+
+**Do NOT call `end-loop`** after creating a PR. That ends ALL remaining iterations. It's a rare escape hatch for when work is stalling across multiple iterations.
+
+## Command Reference
+
+```bash
+# Task lifecycle
+kspec task start @ref
+kspec task note @ref "..."
+kspec task submit @ref
+kspec task complete @ref --reason "..."
+kspec task block @ref --reason "..."
+
+# Task discovery
+kspec tasks ready
+kspec tasks ready --eligible
+kspec task get @ref
+
+# Validation
+kspec validate
+kspec validate --alignment
+
+# Session context
+kspec session start
+```
+
+## Integration
+
+- **`/kspec:writing-specs`** — Create specs before deriving tasks
+- **`/kspec:plan`** — Plans create specs that become tasks
+- **`/kspec:review`** — Review checks AC coverage and code quality
+- **`/kspec:observations`** — Capture friction found during task work
+- **`/kspec:reflect`** — Session reflection after completing tasks

package/templates/skills/triage/SKILL.md

@@ -0,0 +1,199 @@
+# Triage
+
+Systematically process items: inbox, observations, or automation eligibility.
+
+## Focus Modes
+
+Use `/triage <mode>` to focus on a specific area:
+
+| Mode | Purpose | Documentation |
+|------|---------|---------------|
+| `inbox` | Process inbox items → specs/tasks | [docs/inbox.md](docs/inbox.md) |
+| `observations` | Process pending observations | [docs/observations.md](docs/observations.md) |
+| `automation` | Assess task automation eligibility | [docs/automation.md](docs/automation.md) |
+
+Without a mode, follow the full triage session pattern below.
+
+## Core Concept: Record → Act
+
+Triage uses a **two-step pattern** that separates decision-making from execution:
+
+1. **Record** the decision (what to do + why)
+2. **Act** on the decision (execute it)
+
+This separation enables review, override, and audit trails.
+
+```bash
+# Step 1: Record what you want to do
+kspec triage record @inbox-ref --action promote --reasoning "Clear feature request"
+
+# Step 2: Execute the decision
+kspec triage act @triage-ref
+```
+
+### Actions
+
+| Action | What `act` does |
+|--------|-----------------|
+| `promote` | Creates task from inbox item snapshot |
+| `delete` | Deletes the inbox item |
+| `defer` | Records deferral, no side effect |
+| `spec-gap` | Creates observation tagged spec-gap |
+| `duplicate` | Deletes the inbox item |
+
+### Lifecycle
+
+```
+record (with action)     override (optional)     act
+    → triaged        →       triaged         →  acted_on
+```
+
+## Full Session Pattern
+
+1. **Get context**
+   ```bash
+   kspec session start --full
+   kspec inbox list
+   kspec meta observations --pending-resolution
+   kspec tasks assess automation
+   ```
+
+2. **Present overview to user**
+   - Inbox items by category
+   - Pending observations by type
+   - Unassessed tasks needing triage
+
+3. **Ask which focus area**
+   - Inbox items
+   - Observations
+   - Automation eligibility
+
+4. **Process that focus area**
+   - Use the relevant sub-document for guidance
+
+5. **Repeat or stop** when user indicates
+
+## Quick Start by Mode
+
+### `/triage inbox`
+
+Process inbox items using the record → act pattern. See [docs/inbox.md](docs/inbox.md).
+
+```bash
+kspec inbox list
+
+# Interactive mode: triage all untriaged items one by one
+kspec triage start
+
+# Or record decisions individually
+kspec triage record @ref --action promote --reasoning "Clear feature need"
+kspec triage record @ref --action delete --reasoning "Outdated"
+kspec triage record @ref --action defer --reasoning "Not ready yet"
+kspec triage record @ref --action spec-gap --reasoning "Missing spec coverage"
+kspec triage record @ref --action duplicate --reasoning "Covered by @other-ref"
+
+# Then execute recorded decisions
+kspec triage act @triage-ref
+kspec triage act @triage-ref --dry-run  # Preview first
+```
+
+### `/triage observations`
+
+Process pending observations. See [docs/observations.md](docs/observations.md).
+
+```bash
+kspec meta observations --pending-resolution
+# For each: resolve, promote to task, or leave
+kspec meta resolve @ref "Resolution notes"
+kspec meta resolve @ref1 @ref2 "Batch resolution"
+kspec meta observations promote @ref --title "..."
+```
+
+### `/triage automation`
+
+Assess task automation eligibility. See [docs/automation.md](docs/automation.md).
+
+```bash
+kspec tasks assess automation
+# Review criteria, fix issues, or mark status
+kspec task set @ref --automation eligible
+kspec task set @ref --automation needs_review --reason "..."
+```
+
+## Triage Commands Reference
+
+### Recording Decisions
+
+```bash
+# Record a triage decision for an inbox item
+kspec triage record @inbox-ref --action <action> --reasoning "why"
+kspec triage record @inbox-ref --action promote --reasoning "..." --evidence @spec-ref
+kspec triage record @inbox-ref --action promote --reasoning "..." --json  # JSON output
+
+# Interactive mode: presents untriaged items one at a time
+kspec triage start
+# Ctrl+C preserves all previously committed records
+```
+
+### Reviewing Decisions
+
+```bash
+# List all triage records
+kspec triage list
+kspec triage list --status triaged         # Only pending execution
+kspec triage list --status acted_on        # Already executed
+kspec triage list --action promote         # Filter by action type
+
+# Get details of a specific record
+kspec triage get @triage-ref
+
+# Export for agent handoff
+kspec triage export --format context       # Markdown context blocks
+kspec triage export --format json          # Full JSON array
+```
+
+### Executing and Overriding
+
+```bash
+# Execute a recorded decision
+kspec triage act @triage-ref
+kspec triage act @triage-ref --dry-run     # Preview without executing
+
+# Override a previous decision (preserves audit trail)
+kspec triage override @triage-ref --action defer --reasoning "Not ready yet"
+# Then act on the updated decision
+kspec triage act @triage-ref
+```
+
+## Bulk Operations with Batch
+
+When processing many items at once, use `kspec batch`:
+
+```bash
+# Record multiple triage decisions atomically
+kspec batch --commands '[
+  {"command":"triage record","args":{"ref":"@ref1","action":"delete","reasoning":"Stale"}},
+  {"command":"triage record","args":{"ref":"@ref2","action":"delete","reasoning":"Duplicate"}},
+  {"command":"triage record","args":{"ref":"@ref3","action":"promote","reasoning":"Clear scope"}}
+]'
+```
+
+Use `--dry-run` to preview. See `/kspec:help` for full batch documentation.
+
+## Key Principles
+
+- **Record before act** - Separate decisions from execution for audit trail
+- **Ask one question at a time** - Don't batch decisions
+- **Spec before task** - Fill spec gaps before creating tasks
+- **AC is required** - Specs without acceptance criteria are incomplete
+- **Use CLI, not YAML** - All changes through kspec commands
+- **Delete freely** - Outdated items should go
+
+## Progress Tracking
+
+At session end, provide summary:
+- Items triaged (recorded decisions)
+- Actions executed (promoted, deleted, deferred, spec-gap, duplicate)
+- Tasks created/updated
+- Observations resolved
+- Remaining items

package/templates/skills/triage/docs/automation.md

@@ -0,0 +1,120 @@
+# Automation Triage
+
+Assess and prepare tasks for automation eligibility. Goal: make tasks self-contained so they can be automated.
+
+## Philosophy
+
+- **Eligible is the goal** - Manual-only should be the exception
+- **Criteria are for visibility** - Help identify what's missing, not auto-approve
+- **Fix issues, don't just assess** - Guide toward making tasks automatable
+
+## Eligibility Criteria
+
+A task is ready for automation when:
+1. Has `spec_ref` pointing to resolvable spec
+2. Spec has acceptance criteria (testable outcomes)
+3. Task type is not `spike` (spikes output knowledge, not code)
+
+**Having spec + ACs is necessary but not sufficient** - you must also verify the spec is appropriate and ACs are adequate for the task.
+
+**Task type does not determine automation eligibility.** Any non-spike task can be `automation: eligible` — including design tasks, refactoring tasks, documentation tasks, etc. The `automation` field is the definitive source. Once a task is marked `eligible`, do not second-guess it based on type, title, or description.
+
+## Workflow
+
+### 1. Get Assessment Overview
+
+```bash
+# Show unassessed pending tasks with criteria status
+kspec tasks assess automation
+
+# See what auto mode would change
+kspec tasks assess automation --auto --dry-run
+```
+
+### 2. Process Each Task
+
+For each task shown:
+
+**If spike:**
+- Mark `manual_only` - spikes are inherently human work
+- `kspec task set @ref --automation manual_only --reason "Spike - output is knowledge"`
+
+**If missing spec_ref or no ACs:**
+- Ask: "Fix now or mark for later?"
+- **Fix now:**
+  1. Create or find appropriate spec: `kspec item add --under @parent --title "..."`
+  2. Add acceptance criteria: `kspec item ac add @spec --given "..." --when "..." --then "..."`
+  3. Link task to spec: `kspec task set @ref --spec-ref @spec`
+  4. Re-assess and mark eligible if appropriate
+- **Mark for later:**
+  - `kspec task set @ref --automation needs_review --reason "Missing spec - needs spec creation"`
+
+**If has spec + ACs:**
+- Review for eligibility:
+  - Is the spec appropriate for this task?
+  - Are the ACs adequate and testable?
+  - Does the task have sufficient context?
+- If yes: `kspec task set @ref --automation eligible`
+- If no: Fix issues or mark `needs_review` with specific reason
+
+### 3. Batch Processing with Auto Mode
+
+For fast triage of obvious cases:
+
+```bash
+# Apply auto mode (spikes → manual_only, missing → needs_review)
+kspec tasks assess automation --auto
+
+# Then manually review the "review_for_eligible" tasks
+kspec tasks ready --unassessed
+```
+
+Auto mode is conservative:
+- Spikes → `manual_only`
+- Missing spec/ACs → `needs_review`
+- Has spec + ACs → **NOT auto-marked** (requires review)
+
+## Quick Commands
+
+```bash
+# Assessment
+kspec tasks assess automation              # Show unassessed with criteria
+kspec tasks assess automation @ref         # Single task
+kspec tasks assess automation --all        # Include already-assessed
+kspec tasks assess automation --auto       # Apply obvious cases
+kspec tasks assess automation --dry-run    # Preview changes
+
+# Setting automation status
+kspec task set @ref --automation eligible
+kspec task set @ref --automation needs_review --reason "Why"
+kspec task set @ref --automation manual_only --reason "Why"
+kspec task set @ref --no-automation        # Clear to unassessed
+
+# Filtering tasks
+kspec tasks ready --unassessed             # Tasks needing assessment
+kspec tasks ready --eligible               # Automation-ready tasks
+kspec tasks ready --needs-review           # Tasks needing human triage
+```
+
+## Assessment Output
+
+```
+@task-slug  "Task title"
+  spec_ref:     ✓ @feature-slug
+  has_acs:      ✓ 3 acceptance criteria
+  not_spike:    ✓ type: task
+  → review_for_eligible (verify spec/AC adequacy)
+```
+
+| Recommendation | Meaning | Auto Mode Action |
+|----------------|---------|------------------|
+| `review_for_eligible` | Passes criteria, needs review | No change (manual review) |
+| `needs_review` | Missing spec or ACs | Sets `needs_review` with reason |
+| `manual_only` | Spike task | Sets `manual_only` |
+
+## Key Principles
+
+- **CLI doesn't auto-mark eligible** - Requires agent/human review
+- **Agents CAN mark eligible** - When reviewing based on user instruction
+- **Add notes when setting status** - Document the "why"
+- **Re-assess after fixes** - After adding spec/ACs, check again