knowns-dev/kn-go

Go Mode — Full Pipeline Execution

Run the entire SDD pipeline from an approved spec: generate tasks → plan each → implement each → verify → commit. No manual review gates between steps.

Announce: "Using kn-go for spec [name]."

Core principle: SPEC APPROVED → GENERATE TASKS → PLAN → IMPLEMENT ALL → VERIFY → COMMIT.

When to Use

• User has an approved spec and wants to execute everything in one shot
• User says "run all", "go mode", "execute everything", or similar
• The spec is already approved (tag: spec, approved)

When NOT to Use

• Spec is still draft — redirect to /kn-spec first
• User wants to review each task individually — use /kn-plan <id> + /kn-implement <id>
• Spec has unresolved open questions — resolve them first

Inputs

• Spec path: specs/<name> (from $ARGUMENTS or ask user)
• Optional: --dry-run to preview tasks without executing

Process

Complete these phases in order. Do not skip phases.

---

Phase 1: Validate Spec

mcp_knowns_docs({ "action": "get", "path": "specs/<name>", "smart": true })

Check:

• Tags include approved — if not, STOP: "Spec not approved. Run /kn-spec <name> first."
• Has Acceptance Criteria — if empty, STOP: "Spec has no ACs."
• No unresolved open questions marked as blocking

mcp_knowns_validate({ "entity": "specs/<name>" })

If validation errors → fix or report before continuing.

---

Phase 2: Generate Tasks

Parse spec for requirements and generate tasks. Same logic as kn-plan --from @doc/specs/<name> but skip the approval gate.

mcp_knowns_tasks({ "action": "create", "title": "<requirement title>",
  "description": "<from spec>",
  "spec": "specs/<name>",
  "fulfills": ["AC-1", "AC-2"],
  "priority": "medium",
  "labels": ["from-spec", "go-mode"]
})

Add implementation ACs per task:

mcp_knowns_tasks({ "action": "update", "taskId": "<id>",
  "addAc": ["Step 1", "Step 2", "Tests"]
})

Report: "Created X tasks from spec. Starting implementation..."

---

Phase 3: Plan + Implement Each Task

Loop through all generated tasks in dependency order (foundational first, dependent last).

For each task:

3a. Take ownership + plan

mcp_knowns_tasks({ "action": "update", "taskId": "<id>",
  "status": "in-progress"
})
mcp_knowns_time({ "action": "start", "taskId": "<id>" })

• Research context: follow refs, search related docs/memories, check templates
• Use search for discovery first. If a task/spec needs assembled execution context, use mcp_knowns_search({ "action": "retrieve", "query": "<keywords>" }) before drafting or executing the plan. Fall back to CLI knowns retrieve "<keywords>" --json if MCP is unavailable.
• Draft and save plan directly (no approval gate)

mcp_knowns_search({ "action": "search", "query": "<task keywords>", "type": "memory" })

mcp_knowns_tasks({ "action": "update", "taskId": "<id>",
  "plan": "1. Step one\n2. Step two\n3. Tests"
})

3b. Implement

• Work through plan steps
• Check ACs as completed
• Run tests/lint/build after each task

mcp_knowns_tasks({ "action": "update", "taskId": "<id>",
  "checkAc": [1, 2, 3],
  "appendNotes": "Implemented: brief summary"
})

3c. Complete task

mcp_knowns_time({ "action": "stop", "taskId": "<id>" })
mcp_knowns_tasks({ "action": "update", "taskId": "<id>",
  "status": "done"
})

3d. Quick validate

mcp_knowns_validate({ "entity": "<id>" })

If errors → fix before moving to next task.

Progress report between tasks:

"✓ Task X/Y done: [title]. Continuing..."

---

Phase 4: Full Verification

After all tasks complete:

mcp_knowns_validate({ "scope": "sdd" })

Report SDD coverage:

SDD Coverage Report
═══════════════════
Spec: specs/<name>
Tasks: X/X complete (100%)
ACs: Y/Z verified

If coverage < 100% → identify gaps and fix.

Also run project-level checks:

# Build/test/lint — adapt to project
go build ./...
go test ./...

---

Phase 5: Commit

Stage all changes and commit with a single conventional commit:

git add -A
git diff --staged --stat

Generate commit message:

feat(<scope>): implement <spec-name>

- Task 1: <title>
- Task 2: <title>
- ...
- All ACs verified via SDD

This is the ONE gate in go mode — ask user before committing:

Pipeline complete. X tasks done, SDD verified.

Ready to commit:

feat(<scope>): implement <spec-name>

Proceed? (yes/no/edit)

---

Context Budget

If context exceeds ~60% during implementation:

1. Finish the current task
2. Commit completed work so far
3. Report progress and remaining tasks
4. Suggest: "Run /kn-go specs/<name> again to continue remaining tasks."

The skill will detect already-done tasks and skip them on re-run.

---

Re-run Behavior

When invoked on a spec that already has tasks:

1. List existing tasks linked to the spec
2. Filter to todo and in-progress only
3. Skip done tasks
4. Continue from where it left off

mcp_knowns_tasks({ "action": "list", "spec": "specs/<name>" })

---

Error Handling

• Build/test fails during a task: Fix the error, re-run tests. If unfixable, mark task as blocked, append notes, continue to next task.
• Spec has conflicting requirements: STOP and ask user to clarify.
• Task depends on blocked task: Skip and report at the end.

---

Shared Output Contract

Required order for the final user-facing response:

1. Goal/result — state what was completed across the full pipeline run.
2. Key details — tasks completed, tasks blocked, SDD coverage, build/test status.
3. Next action — commit confirmation, or remaining work if interrupted.

For kn-go, the key details should cover:

• total tasks created and completed
• any blocked or skipped tasks
• SDD coverage percentage
• build/test/lint status
• commit proposal

---

Dry Run Mode

With --dry-run:

• Phase 1: validate spec ✓
• Phase 2: generate task preview (don't create) ✓
• Phase 3-5: skip

Show what would be created and ask user to confirm before running for real.

---

Checklist

• [ ] Spec is approved
• [ ] Spec validated (no broken refs)
• [ ] Tasks generated with fulfills mapping
• [ ] Each task: planned → implemented → ACs checked → validated → done
• [ ] SDD verification passed
• [ ] Build/test/lint passed
• [ ] User approved commit
• [ ] Commit created

Red Flags

• Running on a draft spec
• Skipping task validation between tasks
• Not checking ACs before marking done
• Committing without user approval
• Ignoring build/test failures
• Not reporting progress between tasks
• Continuing past context budget limit without checkpointing