do
Skill0 starsChanged 6 months agoMITView on GitHub
Use when the user wants to run the complete cycle on a ticket or description (todo → dev → verify-dev → build → test, with automatic correction loops). Triggers: "do ticket BUG-5 from A to Z", "/do CDP-8.9", "handle this problem entirely", "resolve and validate".
In this repository14 other files it gives its agents.
- buildSkill
- devSkill
- doc-updateSkill
- huntSkill
- knowledgeSkill
- pushSkill
- questionsSkill
- refactorSkill
- security-reviewSkill
- synthesizeSkill
- testSkill
- todoSkill
- verify-devSkill
- verify-todoSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 5 are distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
After execution, read tasks/todo.md and locate the relevant ticket (the most recently added, or the one matching the provided identifier)
- Distinctive: line 33
Wait for the complete end of development
- Distinctive: line 43
Read the produced report carefully
- Very specific: line 54
Read tasks/todo.md and list all ❌ TODO tickets in the format [ticket]-fix-N
- Distinctive: line 66
Read the report verdict:
- Very specific: line 144
Extract the precise ticket: never guess the number, always read tasks/todo.md after /todo
- Distinctive: line 146
One invocation at a time: do not parallelize skills, each step depends on the previous one
- Distinctive: line 147
Stop cleanly: after 3 cycles per phase (verify, build, test) if validation is not reached, and report the exact state
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Full Orchestrator — /do
You are an orchestrator that automates the full cycle: ticket → development → verification → correction, until final validation.
Argument
ARGUMENTS — existing ticket identifier (e.g., BUG-5, CDP-8.9) or free-form description of a problem/feature to create.
Orchestration workflow — MANDATORY, NO STEP CAN BE SKIPPED
Step 1 — Ticket creation / resolution via /todo
⚠️ ALWAYS invoke
/todo— whether the argument is an existing ticket or a free-form description.
Invoke the /todo skill with the argument as provided.
/todoknows how to handle both cases: existing ticket (it finds it and can complete/rephrase it) or free-form description (it creates a new ticket)- After execution, read
tasks/todo.mdand locate the relevant ticket (the most recently added, or the one matching the provided identifier) - Extract its exact number as it appears in the file (e.g.,
CDP-11,BUG-5,ENV-3.1) - Move to Step 2
Step 2 — Development via /dev
⚠️ MANDATORY — This step must be executed after every ticket resolution.
Invoke the /dev skill with the ticket number extracted in Step 1.
- Wait for the complete end of development
- Note all the modified files
→ Once /dev is done: move IMMEDIATELY to Step 3. DO NOT STOP.
Step 3 — Logical verification via /verify-dev
⚠️ MANDATORY — This step must be executed immediately after every
/dev.
Invoke the /verify-dev skill with the same ticket number.
- Read the produced report carefully
- Determine the overall verdict:
- ✅ VALIDATED → move to Step 4 (build)
- ⚠️ VALIDATED WITH RESERVATIONS or ❌ REJECTED →
-fix-Ntickets were created → move to Step 3b
→ Only move to Step 4 if the verdict is ✅ VALIDATED with no CRITICAL/MEDIUM tickets open.
Step 3b — Correction loop (if verify-dev found issues)
⚠️ MANDATORY if
-fix-Ntickets are created — Repeat until validation.
- Read
tasks/todo.mdand list all❌ TODOtickets in the format[ticket]-fix-N - For each correction ticket (in order), invoke
/devwith its exact identifier - Once all fixes are implemented, invoke
/verify-devwith the original ticket (not the fix) - Repeat from point 1 until obtaining ✅ VALIDATED or ⚠️ VALIDATED WITH RESERVATIONS without new CRITICAL/MEDIUM tickets
Safety limit: maximum 3 verification→correction cycles. After 3 cycles with residual CRITICAL/MEDIUM problems, stop and report the state instead of looping indefinitely.
Step 4 — Build & quality via /build
⚠️ MANDATORY — This step must be executed after validation by
/verify-dev. The code may be logically correct but not compile / break the lint / break automated tests.
Invoke the /build skill (without argument, it auto-detects the stack).
- Read the report verdict:
- ✅ DELIVERABLE → move to Step 5 (functional tests)
- ⚠️ DELIVERABLE WITH RESERVATIONS → non-blocking warnings → move to Step 5 (but note the warnings in the final report)
- ❌ BLOCKED → broken build / TypeScript error / lint in error / failed unit test → move to Step 4b
→ Only move to Step 5 if the verdict is ✅ or ⚠️ (never ❌).
Step 4b — Build correction loop
⚠️ MANDATORY if
/buildis ❌ BLOCKED.
- For each error
[BUILD-N]in the report, create a ticket[ticket]-build-fix-Nintasks/todo.md(title = nature of the error, link to the offending file:line) - Invoke
/devon each-build-fix-Nticket in order - Once all fixes are implemented, invoke
/buildagain - Repeat until obtaining ✅ DELIVERABLE or ⚠️ DELIVERABLE WITH RESERVATIONS
Safety limit: maximum 3 build→correction cycles. After 3 cycles with residual errors, stop and report the state.
Step 5 — Functional validation via /test
⚠️ MANDATORY — This step must be executed after a green build. A passing build does not prove that the feature works —
/testvalidates runtime behavior.
Invoke the /test skill with a short description of the behavior to validate, derived from the original ticket (e.g., /test "project creation via POST /projects returns 201 with id").
- Read the report verdict:
- ✅ VALIDATED → cycle finished, move to Step 6
- ⚠️ PARTIAL → some TCs pass, others fail → move to Step 5b
- ❌ FAILED → the feature doesn't work → move to Step 5b
→ Only move to Step 6 if the verdict is ✅ VALIDATED.
Step 5b — Functional test correction loop
⚠️ MANDATORY if
/testis not ✅ VALIDATED.
- For each ❌ TC in the report, create a ticket
[ticket]-test-fix-Nintasks/todo.md(title = expected vs obtained behavior) - Invoke
/devon each-test-fix-Nticket in order - Once all fixes are implemented, re-run
/build(Step 4) to make sure no regression was introduced, then/testagain - Repeat until obtaining ✅ VALIDATED
Safety limit: maximum 3 test→correction cycles. After 3 cycles with failing TCs, stop and report the state.
Step 6 — Final report
Produce a cycle-end summary:
## /do — Cycle complete
**Problem handled**: [the original argument]
**Ticket**: [number and title]
**Dev/verify cycles**: [N]
**Build cycles** : [N]
**Test cycles** : [N]
**Final verdict**: ✅ DELIVERED AND VALIDATED / ⚠️ DELIVERED WITH RESERVATIONS / ❌ BLOCKED
**Modified files**:
- list of touched files
**Applied corrections** (if applicable):
- fix-1 (verify): [short description]
- build-fix-1 : [short description]
- test-fix-1 : [short description]
**Final build** : ✅ DELIVERABLE / ⚠️ with warnings (list them)
**Final tests** : N/N PASS (or: N PASS / M FAIL if ⚠️)Absolute rules
- Always start with
/todo— even if the argument is an existing ticket, no exception - Chain the 5 steps mandatorily:
/todo→/dev→/verify-dev→/build→/test. None can be skipped. - Never stop after
/dev—/verify-devMUST be invoked immediately after - Never stop after
/verify-dev✅ —/buildMUST be invoked next, the code may be logically correct but not compile - Never stop after
/build✅ —/testMUST be invoked next, a green build does not prove the feature works - Never stop on a failure (verify ❌, build ❌, test ❌/⚠️) if
-fix-N/-build-fix-N/-test-fix-Ntickets can be created — the loop MUST continue - After each post-test correction: re-run
/buildthen/testto detect regressions - Extract the precise ticket: never guess the number, always read
tasks/todo.mdafter/todo - Pass the right identifier to
/dev//verify-dev: the exact format as it appears in the TODO - One invocation at a time: do not parallelize skills, each step depends on the previous one
- Stop cleanly after 3 cycles per phase (verify, build, test) if validation is not reached, and report the exact state
$ARGUMENTS
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.

