Value demo — cold run receipt (REQ-VP2)

Mock adapter — this receipt uses [agent] adapter = "mock" for a fast, CI-reproducible proof. For a live-agent dogfood run with the same verify gate, see final-report § D1.

Reproducible evidence that Kelix completes a well-specified goal unattended and returns verified commits. Captured from samples/value-demo/ on 2026-07-02.

Reproduce

From a Kelix checkout with an editable install (pip install -e .):

export KELIX_ROOT="$(pwd)"
export KELIX_VENV="$KELIX_ROOT/.venv/bin"

# Clean nested repo (do not run inside the monorepo root)
DEMO_TMP="$(mktemp -d /tmp/value-demo-XXXXXX)"
cp -r samples/value-demo/. "$DEMO_TMP/"
cd "$DEMO_TMP"

/usr/bin/time -p bash run-demo.sh 2>&1 | tee transcript.txt

Or from the bundled scaffold (initializes git on first run):

cd samples/value-demo
./run-demo.sh

Receipt paths after a successful run:

Artifact Path
Run metadata .kelix/runs/<run_id>/run.json
Retrospective .kelix/runs/<run_id>/retrospective.md
Per-iteration logs .kelix/runs/<run_id>/iter-NNN.log
Verified branch kelix/run-<run_id> (git worktree under .kelix/worktrees/)
Outcome ledger .kelix/memory/loop-metrics.json

Goal

From samples/value-demo/GOAL.md:

Gold in, diamonds out. Build a tiny stdlib-only calculator module that Kelix can complete in five one-iteration tasks.

Outcome: calc.py with add, sub, mul, div (zero-safe), and a CLI entrypoint. Every function has a pytest assertion in tests/test_calc.py.

Acceptance: python3 -m pytest -q exits 0 after all backlog tasks are verified done.

Planning interview

This scaffold ships with a pre-written roadmap and five status: ready backlog tasks, so run-demo.sh skips kelix plan when .kelix/roadmap.md contains Milestone M1. No QUESTIONS.md is produced.

For a live-agent cold run from scratch, delete .kelix/roadmap.md and .kelix/backlog.md, then re-run; kelix plan --goal-file GOAL.md will interview (or write .kelix/phases/<slug>/QUESTIONS.md when non-interactive).

Promote step

Before kelix run, tasks must be status: ready. The bundled scaffold already has five ready owner tasks (T1–T5); no manual promotion is required for the mock demo. After a fresh kelix plan, change status: proposed → status: ready for each task you want the loop to pick up.

Captured run — 20260702-184413

Metric Value
Run id 20260702-184413
Branch kelix/run-20260702-184413
Adapter mock (mockdir/001.sh … 005.sh)
Iterations 5 (5 verified, 0 failures)
Wall clock 3.73 s (/usr/bin/time -p)
Verify gate pytest -q exit 0 (6 passed)

Iteration summary

# Task Rationale Duration Verified
1 T1 add() add() is the highest-priority ready task 0.6 s yes
2 T2 sub() sub() is the highest-priority ready task 0.6 s yes
3 T3 mul() mul() is the highest-priority ready task 0.6 s yes
4 T4 div() div() with zero guard is the highest-priority ready task 0.6 s yes
5 T5 cli cli entrypoint is the highest-priority ready task 0.7 s yes (sentinel)

Verified commits (run branch)

b3f296a14f3aa065c050da5bfec0d7e7681a83da  T1: add function
b8d7c4d348d83ba121c8c41c3410285a5d2fa480  T2: sub function
25d10209dededdeaac68d420cf742090190607bf  T3: mul function
a1142bc56e8fb776c7ca85da1ceed7c66ef505c0  T4: div with zero guard
c06766bd9e631f392a596970cc4ea912ce78a284  T5: cli entrypoint

Inspect on the run branch:

git log kelix/run-20260702-184413 --oneline

Verify receipt

From run.json → last_verify_report:

pytest -q  exit 0
......                                                                   [100%]
6 passed in 0.00s

Run-complete output:

◉ run 20260702-184413 finished: completed (5 iterations, 5 verified-done)
◉ verify: pytest -q exit 0
◉ verified commits: b3f296a..., b8d7c4d..., 25d1020..., a1142bc..., c06766b...

Full stdout transcript

◉ initialized: .kelix/phases/README.md
↻ next moves:
    1. describe your goal in GOAL.md
    2. kelix plan --goal-file GOAL.md   (it will interview you)
    3. review the draft, promote tasks to ready
    4. kelix run                        (wake up to verified commits)
Roadmap present — skipping kelix plan.

=== PROMOTE STEP ===
This scaffold ships with status: ready tasks (mock demo needs no edits).
For a live agent run after kelix plan, promote proposed tasks in
.kelix/backlog.md from status: proposed to status: ready before run.
kelix run 20260702-184413: branch=kelix/run-20260702-184413 cap=10
  iter 1: rationale=T1 — add() is the highest-priority ready task progress=True verified=True ok
  iter 2: rationale=T2 — sub() is the highest-priority ready task progress=True verified=True ok
  iter 3: rationale=T3 — mul() is the highest-priority ready task progress=True verified=True ok
  iter 4: rationale=T4 — div() with zero guard is the highest-priority ready task progress=True verified=True ok
  iter 5: rationale=T5 — cli entrypoint is the highest-priority ready task progress=True verified=True ok
  distill: calc.py: not under .kelix/skills/_proposed/
  distill: tests/test_calc.py: not under .kelix/skills/_proposed/
◉ run 20260702-184413 finished: completed (5 iterations, 5 verified-done)
◉ verify: pytest -q exit 0
◉ verified commits: b3f296a14f3aa065c050da5bfec0d7e7681a83da, b8d7c4d348d83ba121c8c41c3410285a5d2fa480, 25d10209dededdeaac68d420cf742090190607bf, a1142bc56e8fb776c7ca85da1ceed7c66ef505c0, c06766bd9e631f392a596970cc4ea912ce78a284
real 3.73
user 1.71
sys 0.65

Retrospective excerpt

From .kelix/runs/20260702-184413/retrospective.md:

- status: **completed**
- iterations: 5 (5 verified, 0 failures)
- branch: `kelix/run-20260702-184413`

## Iterations
- 1: T1 — add() is the highest-priority ready task -> verified
- 2: T2 — sub() is the highest-priority ready task -> verified
- 3: T3 — mul() is the highest-priority ready task -> verified
- 4: T4 — div() with zero guard is the highest-priority ready task -> verified
- 5: T5 — cli entrypoint is the highest-priority ready task -> verified

What this proves