Skip to content
Agentmart
Submit a tool

submission-qa

Verification methodology and gate checklist for the tdmcp Connectors Directory submission — run the docs build, build and inspect the .mcpb bundle, sweep for…

Skill by hybridlabor-api

IndexedApache-2.0★ 6
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/hybridlabor-api-submission-qa

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Verification methodology and gate checklist for the tdmcp Connectors Directory submission — run the docs build, build and inspect the .mcpb bundle, sweep for stale .dxt references, and cross-check the form-answer draft against the real requirements. Use to validate the submission package before a human submits, by executing commands and quoting evidence, not by trusting that work was done.

Use cases: testing

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/submission-qa

What's inside

Skill name
submission-qa
Description
Verification methodology and gate checklist for the tdmcp Connectors Directory submission — run the docs build, build and inspect the .mcpb bundle, sweep for stale .dxt references, and cross-check the form-answer draft against the real requirements. Use to validate the submission package before a human submits, by executing commands and quoting evidence, not by trusting that work was done.
License
Apache-2.0
Instructions
46 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Submission QA — verify by executing

The cost of a miss is a ~2-week rejection cycle. So every gate is checked by
**running something and quoting the output**, never by reading a builder's claim.
"Looks complete" is not a result; a command transcript is.

## Gate checklist

Produce a table `gate | PASS/FAIL | evidence`. Gates:

| Gate | How to verify |
|------|---------------|
| Privacy page exists + linked | Open the page file(s); confirm the nav entry in `docs/.vitepress/config.ts` points to its clean URL. Confirm the page actually states the data-handling story (not a stub). |
| Docs build | `npm run docs:build` exits 0. VitePress fails on dead internal links, so this also catches a bad nav link. |
| Bundle builds | `npm run build` then `npm run build:mcpb` (or the renamed script). Confirm `tdmcp.mcpb` is emitted. |
| Bundle is valid | `unzip -l tdmcp.mcpb` shows `manifest.json` at root + `dist/`. If `npx --yes @anthropic-ai/mcpb validate` exists, run it on the manifest and capture the result. |
| No stale refs | `grep -rn "\.dxt\|build:dxt" docs/ scripts/ README.md package.json .github/ 2>/dev/null` (exclude `dist/`, `node_modules/`). Every hit must be an intentional legacy note. |
| Annotations intact | `grep -rEo "readOnlyHint|destructiveHint" src/tools | wc -l` is non-zero and matches expectation; spot-check a read tool has no `destructiveHint` and a write tool does. (#1 rejection cause — confirm no regression.) |
| Form completeness | Every field in `00_submission-spec.md`'s field map has an answer in `02_form-answers.md` or an explicit `NEEDS HUMAN INPUT`. List the human inputs. |
| Manifest sane | `node -e "JSON.parse(require('fs').readFileSync('dxt/manifest.json'))"` parses; `manifest_version`, `name`, `version`, `server.entry_point` present. |

## Report format

Write `_workspace/03_qa-report.md`:
1. **Gate table** (above) with command output as evidence.
2. **Blocking issues** — anything FAIL. Each with the exact file:line or command
   and what the owner (docs-author / bundle-engineer) must change.
3. **Non-blocking nits** — cosmetic / nice-to-have.
4. **Needs human before submit** — every `NEEDS HUMAN INPUT` (support email,
   privacy contact, test account, new release to publish the `.mcpb` asset).

## Principles

- You need to **run scripts**, so operate with `general-purpose` capability, not a
  read-only profile.
- **Incremental:** when a gate fails, report it precisely and (on re-run)
  re-execute only the affected check rather than the whole suite.
- **Don't fix.** Independent verification is your value — hand fixes back to the
  owning agent. If you fix and verify your own fix, the check is no longer
  independent.
- **Cross the boundary.** The high-value check isn't "does the file exist" — it's
  "does the form answer match what the page/manifest/repo actually says". E.g.
  the privacy URL in `02_form-answers.md` must resolve to the page docs-author
  actually created; the tool list in the form must match `src/tools/**`.

From hybridlabor-api/aos, Apache-2.0.

More from hybridlabor-api/aos

Signals

Stars
6
Forks
0
Open issues
19
Last commit (this folder)
Aug 13, 2026 (2 months ago)
Last release
Oct 14, 2026 (15 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
Apache-2.0
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from hybridlabor-api

Cite this page

submission-qa skill. Agent Mart. https://agentmart.dev/skills/hybridlabor-api-submission-qa (last updated 2026-10-24)