Session Journal — 2026-07-06¶
SPEC-153 — build phase (--draft pd-3 doc-backflow)¶
Activity — build complete, behavioral spot-check passed¶
What was done: Shipped SPEC-153 --draft mode on /vt-c-user-manual-update as a prose-wired
skill extension + discoverability-guide edit + structural bats drift-guard (tests/spec-153/).
No runtime code — user-manual-update is an agent-instruction skill.
Result:
- bash tests/spec-153/run.sh → 14/14 GREEN.
- bash tests/spec-135/run.sh → 11/11 GREEN (regression: the counted decision-log chain stays
exactly 3 rows; the backflow lives in a separate "Downstream consumers" subsection OUTSIDE the
anchor, so the two drift-guards stay independent).
Behavioral spot-check (why it mattered): the drift-guard only proves contract presence — it
cannot prove an agent following the prose produces a valid draft. Per the plan's Test Strategy I
hand-executed --draft against a 3-iteration booking-flow decision-log sandbox (in scratchpad).
All six success criteria verified:
- SC-1 page-top notice DRAFT — pd-3 iteration, expect changes. present.
- SC-3 no decision entries → clean "no pd-3 decisions to draft from" exit, no file written.
- SC-4 finalized docs/user-manual/docs/features/ surface untouched.
- SC-5 works with no mkdocs scaffold — creates only the self-contained docs/user-manual/drafts/.
- SC-6 no fabricated How-it-works — thin template leaves an explicit TODO, invents nothing.
- SC-7 supersede/overwrite — a single per-feature page reflects the latest iteration (iter-3),
not an accumulation.
Key decision honored (Decision 1 / FR-5): the draft uses a thin decision-log-derived template (What's being built / Why / Files touched), NOT the base acceptance-criteria template — the decision-log has no user stories, so any "How it works" would be fabrication.
Files: plugins/core-standards/skills/user-manual-update/SKILL.md (additive --draft block,
Steps 1–10 byte-unchanged), docs/workflows/product-design/prototype-decision-log.md (guide),
tests/spec-153/ (new drift-guard). Commits 77740a99 (RED), 21e71f1d (skill), 2b4aaee2
(guide), b3655f41 (build gate).
Note: skill-map soft-warning on the SKILL commit is a known false trigger for prose-only SKILL additions (no orchestration-structure change; SPEC-134/135 precedent) — map not regenerated.