verifying-work
Skill1 starsChanged 5 months agoMITView on GitHub
The required verification discipline for the Slack-Claude Bridge — run, test, and observe a change end-to-end before marking it done; if no test/integration path exists, build it autonomously or ask for the human-only inputs needed. Use before claiming any feature or fix is complete, fixed, or passing.
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- bridge-domainSkill
- reviewing-changesSkill
- run-bridgeSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 28
If there is no complete way to exercise the change — no test, no runnable entry point, no mock event, no way to observe the result — then …
- Distinctive: line 31
Build the missing tooling autonomously: Add the unit/integration test, a fixture or mock Slack event, a local harness/script, a fake …
- Distinctive: line 35
Escalate only what genuinely needs a human: If verification truly requires something you cannot create or obtain — real Slack credentials, …
- Distinctive: line 39
Never mark done on assumption: If a step still cannot be verified after building tooling and asking, say exactly which step is unverified …
- Distinctive: line 42
Default posture: _if it can be automated, automate it and verify.
≈ “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
Verifying your work
A change is not done until you have run, tested, and verified it end-to-end. Mandatory, not aspirational. Report evidence (command output, log lines), never bare assertions.
The loop — close it before reporting complete
- Build / type-check —
npm run verifypasses (or the relevant subset:npm run typecheck,npm run lint,npm run test). - Run it — start the bridge and confirm it boots without throwing. See the
run-bridgeskill for how to start it and taillogs/bridge.log. - Observe output — confirm the expected events in the log: Slack connection, event received, workspace routed, Claude invoked, response posted — each prefixed with the mapping identity.
- Integration — exercise the path end-to-end where feasible (a test, a mock Slack event, or a real message in a dev channel) and confirm the response lands in the correct location (channel vs thread, per session scope).
- Report evidence — quote the output / log lines that prove it.
Aligns with the superpowers:verification-before-completion discipline.
No untested "done" — build the verification path if it is missing
If there is no complete way to exercise the change — no test, no runnable entry point, no mock event, no way to observe the result — then creating that path is part of the task:
- Build the missing tooling autonomously. Add the unit/integration test, a fixture or mock Slack event, a local harness/script, a fake workspace, a log assertion — whatever lets the behavior be exercised and observed without manual guesswork. This is expected work, not a follow-up.
- Escalate only what genuinely needs a human. If verification truly requires something you cannot create or obtain — real Slack credentials, a workspace/channel only the user controls, a paid resource, an external approval — stop and ask the user for that specific input or help. Be precise about what you need and why.
- Never mark done on assumption. If a step still cannot be verified after building tooling and asking, say exactly which step is unverified and why — never imply success.
Default posture: if it can be automated, automate it and verify; if it needs a human, ask. "There was no way to test it" is not an acceptable reason to skip verification — making a way to test it is the job.
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.

