qa-answer-review
Skill2 starsChanged 2 days agoMITView on GitHub
Review the prepared answers in a "Q&A" sheet before a customer / go-live / integration meeting: pull the tab (No · Question · Prepared answer · Remark), run every answer through four review lenses in parallel — business, program, data, operation (lenses/<lens>/SKILL.md) — merge the verdicts into one plain note per row and write it into the Remark column. Use WHENEVER the work is "review the Q&A before the meeting", "are these prepared answers safe to say", "ตรวจคำตอบก่อนประชุม", "ลง remark ให้ Q&A", "business review / program review of an answer", or a spreadsheet link that names a Q&A tab. Read-only on source code and databases; writes only the Remark column.
In this repository23 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- new-skillSkill
- agent-kit-costSkill
- agent-kit-prompt-cacheSkill
- agent-kitSkill
- containers-podmanSkill
- containersSkill
- dependency-vuln-explainerSkill
- design-coreSkill
- design-flowSkill
- devops-changeSkill
- devops-network-diagramSkill
- devops-release-scopeSkill
- devopsSkill
- documentation-doc-styleSkill
- documentation-pdf-operationsSkill
- documentationSkill
- poc-spec-driven-builderSkill
- qa-excel-importSkill
- qa-export-evidenceSkill
- qaSkill
- qa-ticket-reportSkill
- thai-mock-namingSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 40
Default Q&A\s*(?:List)?\s*(\S+) matches Q&A List ACME, Q&A ACME.
- Distinctive: line 43
Never edit the Question or Answer columns.
- Very specific: line 96
Write the whole Remark column in one paste / one batch update built from the live column with plan.json applied.
- Distinctive: line 98
Verify by re-export: the Remark column equals the plan, every other cell equals the pre-write copy.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Review a prepared Q&A tab
In a sheet export (.xlsx) with one tab per meeting · Out one Remark per question saying
whether the prepared answer is safe to say, what to change, and what is still open — plus the
evidence trail in <out>/<tab-key>/<DDMMYYYY>/.
The answers are what we will tell a customer or partner in a meeting. The review question is not "is this nice prose" but "if we say this, is it true, is it ours to promise, and can the system back it up today?"
S=skills/qa/answer-review/scripts
# 1. pull: export the sheet as .xlsx (Drive: File → Download → .xlsx, or an MCP download that spills JSON with base64 content)
uv run $S/qa.py pull <sheet.xlsx | download.json> --out work/qa-review \
[--tab 'Q&A List (\S+)'] [--headers "No,Question,Answer,Remark"] # -> work/qa-review/<key>/qa.json
# 2. lenses write work/qa-review/<key>/<DDMMYYYY>/lens-<lens>.json (one agent per lens, in parallel)
# 3. condense the lens notes into final.json by hand, then merge
uv run $S/qa.py merge <key> --out work/qa-review # -> plan.json (old / new per row)
uv run $S/qa.py selfcheck0. The tab
- Header row 1 holds the four columns. The default is
No · Question · Answer · Remark; pass--headerswith the sheet's exact labels (any language).qa.py pullasserts them and stops if a column was renamed. --tabis a regex with one capture group: the group becomes the folder key (TOP,ACME, …). DefaultQ&A\s*(?:List)?\s*(\S+)matchesQ&A List ACME,Q&A ACME.- Remark may already hold team comments on some rows. Those are review input, not noise — every lens reads them, and they stay in the cell.
- Never edit the Question or Answer columns. A better answer goes into Remark as a suggestion; the team owns the wording.
1. Lenses — one agent each, in parallel
| Lens | Asks | Skill |
| business | Is it ours to say? Process, roles, scope, commitments, consistency with the other answers | lenses/business/SKILL.md |
| program | Does the system do this today? Endpoint · screen · released tag | lenses/program/SKILL.md |
| data | Do the fields, keys and units named exist, with those meanings, and do they join? | lenses/data/SKILL.md |
| operation | Network, file transfer, sandbox, auth, monitoring, support — can we run it? | lenses/operation/SKILL.md |
Spawn four read-only agents, each over all rows of qa.json, each told to read its lens
SKILL.md and follow it. Every brief carries, verbatim:
Read-only: no repo edits, no DB writes, no logins, no sheet edits. Repos:
<list them>— reportgit log -1of each. Database:<read-only command>. Input:<out>/<key>/qa.json. Output:<out>/<key>/<DDMMYYYY>/lens-<lens>.json=[{No, row, verdict, evidence, note}], one item per row,verdictone ofOK · ADJUST · WRONG · OPEN · N/A.evidenceis for us (file:line, tag, query + result, doc path).noteis for the meeting team: 1–2 short lines in the team's language, lead with what was found ("checked: …" · "from what we found, … likely …" · "need more info on …"), never "cannot prove" / "no conclusion", never name a tool, script, skill or agent.N/Awhen the row is outside the lens — leavenoteempty then.
For Thai notes follow the kien-thai rules (technical terms stay English, no middle dot ·).
| Verdict | Means | Remark shows |
OK | true, ours to say, backed by the system today | one line, the evidence in words |
ADJUST | right direction, but overclaims, underclaims or is vague | what to change, a better phrase |
WRONG | contradicts the system or the process | what is actually true |
OPEN | depends on a decision not yet made | who decides, and a holding answer |
N/A | outside this lens | nothing |
2. Merge, propose
- Condense by hand into
final.json={No: "- …\n- …"}. The Remark reads like the team's own notes — 1–3 plain-bullets, sorted (what is true → what to say instead → what to ask), no lens labels, no verdicts, no date header. Merge the four lenses into one view; where they disagree, write it as a thing to check, not two answers. qa.py merge <key>writesplan.json: the team's text kept on top, a blank line, our bullets. Our bullets from any earlierfinal.jsonof that key are stripped first, so a re-review replaces them instead of stacking.- Show the owner a short table: No · verdicts per lens · the one-line gist. Rows with
WRONGfirst. Once shown, write — unless they stop it.
3. Write Remark (the fourth column)
Whatever drives the sheet (browser automation, Sheets API, a CSV re-import), keep these checks:
- Re-export the tab, parse it quote-aware, assert the Remark header and that every
plan.jsonrow'soldstill equals the live cell. A mismatch means a teammate edited meanwhile: re-pull and re-merge, don't overwrite. - Write the whole Remark column in one paste / one batch update built from the live column with
plan.jsonapplied; never cell by cell by hand. - Verify by re-export: the Remark column equals the plan, every other cell equals the pre-write copy.
plan.json(old/newper row) is the before/after record — no separate CSV needed.- A large plan does not fit in a script argument: load it from a file.
Traps
- The meeting is about the partner's integration, not our UI. "API" in a question can mean their current push, our external API, or a webhook. Say which one in the note.
- "Already available" needs a tag. Merged on
developis not available to a customer; quote the release that carries it, or say it is not released yet. - Answers lean on each other. A row that offers "3 options: SFTP, Excel, manual" is contradicted by a Remark saying there is no SFTP — the business lens checks the set, not just the row.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

