triage-batch
Skill0 starsChanged 51 days agoMITView on GitHub
Triage a list of Oracle HCM Jira tickets in one run and write your steps and your messages to output/
- Installs packages
In this repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- smoke-checkSkill
- fusion-instanceSkill
- fusion-referenceSkill
- fusion-suiteSkill
- fusion-testSkill
- hcm-architectureSkill
- hcm-configurationSkill
- hcm-debuggingSkill
- jira-ticket-ingestionSkill
- triageSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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
- Distinctive: line 14
If the argument is a file path, pass it through.
- Very specific: line 30
Read skills/hcm-debugging/output-format.md before writing the first brief.
- Very specific: line 44
Read tickets/{KEY}/ticket.md in full and open every image in tickets/{KEY}/images/.
- Very specific: line 51
Write output/tickets/{KEY}.md in the dense format.
- Distinctive: line 53
Do not batch the reading of several tickets before writing: finish one brief, then start the next.
- Distinctive: line 78
Do not wait for the whole list to finish.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Triage every ticket in this list: $ARGUMENTS
Jira stays read-only for the whole command. GET only, nothing written back to any ticket.
1. Ingest the whole list first
If the argument is a file path, pass it through; otherwise pass the keys directly:
node dist/jira_ingest.js --from-file $ARGUMENTS # a file of keys
node dist/jira_ingest.js --keys $ARGUMENTS # keys given inlineBuild first with npm install && npm run build if dist/jira_ingest.js is missing. On an
authentication failure, show the credential instructions the script printed and stop: do not triage
tickets you could not fetch.
Ingest everything before analysing anything. It is one authenticated pass, it fails fast, and it tells you the real size of the run (comment and attachment counts per ticket).
2. Read output-format.md once
Read skills/hcm-debugging/output-format.md before writing the first brief. Every brief in the run
follows it exactly, because npm run actions parses the frontmatter, the DO NOW table and the
MESSAGE section, and rejects a brief that drifts.
Two things in that file decide the shape of every brief. First, my_move: investigate means there
are steps to run, send means the analysis is finished and the message is the deliverable, wait
means someone else moves first. Second, the scope rule: a row in DO NOW is something the operator
types or clicks himself. Anything owned by Oracle, the client or a colleague belongs in the message
or in NOT MINE.
3. Triage each ticket, one at a time
For each key, in the order given:
- Read
tickets/{KEY}/ticket.mdin full and open every image intickets/{KEY}/images/. In this corpus the error text, the page state and the configuration values live in the screenshots. - Apply the
hcm-debuggingskill end to end: pattern library, evidence, the step 1b split when the ticket holds several problems, classification, research through theoracle-fusion-docsMCP then the web, ranked hypotheses each with one discriminating test and an in-instance verification path, then step 5 (his keystrokes, after the owner filter and the root-cause filter) and step 6 (the full text of the message). - Write
output/tickets/{KEY}.mdin the dense format.
Do not batch the reading of several tickets before writing: finish one brief, then start the next. A half-read ticket produces a confident brief about the wrong problem.
4. Generate the action folder
npm run actionsFix any FORMAT ERROR it reports, then run it again until it is clean.
5. Report back
In the conversation, give only:
- the total: how many steps to run, how much keyboard time, how many messages ready to send
- the first ticket to open and its first step, verbatim from
DO-NOW.md - anything found that changes priorities across the queue (a live risk, a ticket that is actually finished, two tickets that share one root cause)
The detail belongs in the files: DO-NOW.md for what he runs, MESSAGES.md for what he sends,
INDEX.md for where the queue stands. The conversation carries the decisions.
Deliver as you go
Do not wait for the whole list to finish. As soon as one ticket's root cause is settled, show its message and the Jira status to set for that ticket, then move on to the next one. A settled ticket held back until the batch completes is a ticket that could have been closed today and was not.
Where the next fact comes from
Work the evidence ladder from the top, and stop at the first rung that settles the question: the dossier you already have, then the Oracle documentation, then a fact only the client can produce, then a read in DEV1, and only last a change in DEV1.
Documentation research is the first reflex, not a fallback. Quote the page verbatim with its URL when it describes the observed behaviour. Opening an instance is the expensive rung and has to earn its place. Asking the client for an extraction they alone can run is a legitimate answer, not a failure to investigate.
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.

