| name | flotilla-seven-c-walk |
| description | Product XO / dash-walker SEVEN-C evening walk — 390px-first Playwright captures, grade grid 0–2 per C, file a GitHub issue when COMPELLING or CORRECT is below 2, store assets under state/<xo>-walk-<YYYYMMDD>/assets/ (not parades/). Read before a product walk, evening-walk dispatch, or SEVEN-C scorecard. |
| type | skill |
| queries | ["SEVEN-C product walk capture scorecard","evening walk 390px playwright flotilla dash","file issue after walk COMPELLING CORRECT","flotilla-dev-walk assets where to save screenshots"] |
| keywords | ["seven-c-walk","product-walk","390px","scorecard","COMPELLING","CORRECT"] |
| boost | 0.08 |
SEVEN-C product walk — capture + grade + file
Audience: product-owning XOs and dash execution walkers (not CoS parade authors).
Ceremony context: docs/coordinator-runbooks/ceremonies.md (evening walk).
Asset paths (load-bearing)
Walk captures and scorecards live under the roster state tree — not parades/:
<roster-dir>/state/<xo>-walk-<YYYYMMDD>/assets/<view>.png
<roster-dir>/state/<xo>-sevenc-scorecard-<YYYYMMDD>.md
Example (product-dev lane): state/flotilla-dev-walk-20260711/assets/w11-goals-viewport-390.png
parades/<date>/assets/ is for morning parade demo slides only. Product-walk evidence
stays in the walk directory so scorecards and issues cite durable, non-ceremony paths.
Capture recipe (390px first)
- Viewport:
390×844 (phone proxy). Desktop captures are supplementary, not the grade bar.
- Tool: Playwright chromium from a throwaway venv (
python3 -m venv /tmp/pw-venv;
pip install playwright; uses ~/.cache/ms-playwright — no sudo chrome install).
- Per surface:
goto live URL → wait_for_selector on a real element → pageerror hook
must stay zero → screenshot(full_page=True).
- Name files by view: e.g.
goals-viewport-390.png, conversations-390.png, parade-list-390.png.
- Vision grade: COMPELLING requires a seeing agent on the PNGs — the capture desk does not
grade its own screenshots.
Minimal script shape:
from playwright.sync_api import sync_playwright
with sync_playwright() as p:
b = p.chromium.launch()
pg = b.new_page(viewport={"width": 390, "height": 844})
errors = []
pg.on("pageerror", lambda e: errors.append(str(e)))
pg.goto("http://127.0.0.1:<port>/")
pg.wait_for_selector("#real-anchor", timeout=10000)
pg.screenshot(path="assets/goals-viewport-390.png", full_page=True)
b.close()
assert not errors
For broad submatrices, use the checked-in fail-soft runner instead of waiting on
one hard-coded populated-state selector:
python3 .claude/skills/flotilla-seven-c-walk/scripts/walk_run.py \
--base-url http://127.0.0.1:8787 \
--output-dir <roster-dir>/state/<xo>-walk-<YYYYMMDD>/assets
Routes, actions, and state anchors live in walk-manifest.v1.json. The runner
tries current routes before legacy fallbacks, records an absent optional state
as unavailable, continues the remaining captures, and atomically finalizes
walk-run.json. A missing required surface or a page error still makes the run
non-zero, but only after the rest of the matrix and its evidence manifest have
completed. Update the versioned manifest when product routes or state grammar
move; do not copy selectors into an evening-specific script.
Grade grid (0–2 each; cite evidence)
| C | Question |
|---|
| COMPLETE | Did we walk every litmus path we owe? |
| CORRECT | Does the UI tell the truth (no false-healthy chrome)? |
| COMPREHENSIVE | Are the operator's daily questions answerable from what rendered? |
| CALIBRATED | Are staleness/age/session labels honest? |
| CONCISE | 390px: no horizontal scroll; primary actions in thumb reach? |
| COMPELLING | Would the operator want to open this on their phone? (pixels + vision) |
| CONSISTENT | Cross-surface signals agree (banner vs row vs tab)? |
Record DELTA vs the prior scorecard file. TOTAL = sum of seven grades (max 14).
Generated work — file issues, don't narrate
Mechanical gate: when COMPELLING < 2 OR CORRECT < 2, open a GitHub issue in the
product repo before settling. Every other C < 2 should also produce filed work, but those two
are the non-negotiable filing triggers.
Issue acceptance shape (model: flotilla#609)
## Summary
One sentence: what failed the walk and user impact.
## Evidence
- Capture: `state/<xo>-walk-<YYYYMMDD>/assets/<file>.png` (roster-local path)
- Walk: SEVEN-C <ISO timestamp> (`flotilla-dispatch-<hex>` if dispatched)
- Prior: #<issue> if regression
## Acceptance
Concrete, testable UI outcome at 390px (readable label, no false-green, etc.).
## Scope
Named subsystem only — not a drive-by refactor.
Dispatch the owning build desk after filing; link the issue in the scorecard GENERATED WORK
section.
Scorecard footer
End the scorecard with total, delta vs prior, and every filed issue URL. Skip-as-satisfied
(~6h) is allowed only with an explicit backlog note — never a silent skip.
