Spec Kitty

Research Mission Quickstart — Operator Dogfood

This is the operator-runnable sequence that proves a fresh research mission can be created and advanced through the composition path with no MissionRuntimeError. WP06's smoke-evidence.md captures one canonical run of this sequence verbatim. Reviewers should re-run it from a clean shell.

Prerequisites

• A checkout of spec-kitty at HEAD (the worktree containing your changes).
• uv installed and on $PATH.

Set the path to your checkout once for the rest of the session:

SPEC_KITTY_REPO=/path/to/spec-kitty   # or the lane worktree

Step 1 — Create an isolated sandbox

cd /tmp && rm -rf demo-research-smoke && mkdir demo-research-smoke && cd demo-research-smoke
git init -q && git commit --allow-empty -m "init" -q

Expected: a fresh empty git repo, no .kittify/ yet.

Step 2 — Initialize spec-kitty

uv --project $SPEC_KITTY_REPO run spec-kitty init --non-interactive --ai claude

Expected: .kittify/ populated; agent dirs (e.g. .claude/) seeded; no error.

Step 3 — Create a research mission

uv --project $SPEC_KITTY_REPO run spec-kitty agent mission create demo-smoke \
    --mission-type research --json

Expected JSON keys (verbatim sample in smoke-evidence.md):

• result: "success"
• mission_type: "research"
• mission_slug: demo-smoke-<MID8> (e.g. demo-smoke-01KQ4Z73)
• mission_id: a 26-char ULID
• feature_dir: <repo>/kitty-specs/<mission_slug>

No MissionRuntimeError, no Python traceback. The v1 P0 finding (MissionRuntimeError: Mission 'research' not found) is closed as soon as this command returns successfully.

Step 4 — First next (query)

uv --project $SPEC_KITTY_REPO run spec-kitty next \
    --agent claude:opus-4.7:test:operator --mission <mission_slug>

Expected:

[QUERY — no result provided, state not advanced]
  Mission: <mission_slug> @ not_started
  Mission Type: research
  Next step: scoping

Mission Type: research and Next step: scoping confirm the runtime template chain resolved the research template.

Step 5 — Advance scoping (composition path)

uv --project $SPEC_KITTY_REPO run spec-kitty next \
    --agent claude:opus-4.7:test:operator --mission <mission_slug> \
    --result success

The first call issues scoping. Run it again with --result success; the second call drives composition for the scoping step and the response shows:

[STEP] <mission_slug> @ methodology
  Mission Type: research
  Action: methodology

methodology is the next research-native step, proving composition fired and the planner advanced through the research DAG (not the software-dev DAG).

Step 6 — Inspect the invocation trail

ls -la .kittify/events/profile-invocations/

Expected: one <ulid>.jsonl file per composed contract step. Each file contains a paired started + completed record. The started record's action field is scoping (the research-native step ID) and profile_id is researcher-robbie.

cat .kittify/events/profile-invocations/<one>.jsonl

Pass criteria

Criterion	Where it shows up
No MissionRuntimeError for 'research'	Step 3 + Step 4 succeed
Research template resolves	Step 4 prints Mission Type: research
Composition path fires for scoping	Step 5 second call returns methodology
Paired invocation records written	Step 6 trail files have started + completed
Action hint preserved on the trail	Step 6 trail file shows "action": "scoping"

If any of these regress, the v1 P0 finding is open again.