← ClaudeAtlas

acceptance-templatelisted

Make every acceptance item a runnable command with an expected result, quote wording exactly, and add a RISKS section. Use when writing or amending a task's spec or acceptance.
volekerb/hivemill · ★ 33 · AI & Automation · score 75
Install: claude install-skill volekerb/hivemill
# Acceptance template Prose acceptance ("works", "is documented") makes QA guess. Each rule below adds a requirement; none loosens plan-quality. A1 Every acceptance item is `N. <claim> - <runnable command> -> <expected result>`; the result is an exit code, an exact output line, a count, or "empty". A judgement item names the exact file:line or diff hunk to read. Good: `3. tests pass - python3 -m unittest discover -s tests -p test_x.py -> exit 0, "OK"`. Bad: `3. tests pass`. A2 Wording that must survive (rule text, messages, markers) is quoted exactly and checked with `grep -nF '<exact text>' <file>`. Good: `grep -nF 'never call the advisor tool' roles/x.md` -> one hit. Bad: "the advisor rule is kept". A3 "Additions only" or "no other line changes" is a grep: `git diff origin/main -- <file> | grep '^-' | grep -v '^--- a/'` -> empty. Bad: "do not change other lines". A4 Byte and line caps are `wc -c` / `wc -l` commands with the number. A role prompt delta states the headroom (plan-quality P8). Good: `wc -c < roles/pm.md` -> at most 4606 (headroom 1362). Bad: "keep the prompt short". A5 Every spec has a `RISKS:` section: each risk, how it is mitigated, and which acceptance item tests it. A spec without RISKS is not ready. Good: `RISKS: (r1) prompt over budget -> item 2 measures it`. Bad: "RISKS: none". A6 Guards: one blocked and one allowed case through the real entry point (see plan-quality P2; do not restate it), plus a C4 item: remove the fix -> the new