fusion-instance
Skill0 starsChanged 51 days agoMITView on GitHub
Read or change an Oracle Fusion TEST instance directly through the fusion CLI (REST GET, text snapshots of pages, clicks and field entry), never production. Use when the next fact can only come from the test pod itself, such as the roles of a user, the content of a security profile, a retest after a change, or an approved configuration change in the test pod. Not for finding documentation (use fusion-reference and fetch_oracle_page), and not a substitute for the evidence ladder in hcm-debugging.
In this repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- smoke-checkSkill
- fusion-referenceSkill
- fusion-suiteSkill
- fusion-testSkill
- hcm-architectureSkill
- hcm-configurationSkill
- hcm-debuggingSkill
- jira-ticket-ingestionSkill
- triage-batchSkill
- triageSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 2 are distinctive, and 6 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.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 49
Only in the test pod, and only a change the user owns.
- Very specific: line 57
Commit with click "<Save or Submit>" --write.
- Very specific: line 58
Verify by reading back, with a snapshot or REST, then retest under the target role.
- Very specific: line 64
Every path walked with goto and click is evidence about the navigation.
- Very specific: line 68
When a match is ambiguous the tool lists the candidates and stops, never guessing.
- Very specific: line 73
Never production. Do not weaken the host guard, and never set FUSION_ALLOW_UNRECOGNISED_HOST without a written decision from the user.
- Distinctive: line 75
Never a credential, never a sign in on the user's behalf.
- Very specific: line 76
Every --write follows an explicit approval in the conversation, for that change.
≈ “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
Working in the test instance
The instance is the last rung of the evidence ladder: the dossier, then the Oracle documentation, then a fact only the client can produce, then a read in the test pod, and only last a change in it. Open the instance only when the rungs above cannot settle the question.
Setup, once per session
npm run fusion -- statusclosed: runnpm run fusion -- open, then ask the user to sign in in the window that opens.open but not signed in: ask the user to sign in. Never type a user name, a password or an MFA code, never ask for them in the conversation. The tool refuses credential fields anyway.signed in: <url>: go.
The session belongs to whoever signed in, usually an administrator. A test under an administrator proves nothing about a role-specific defect. To retest under a role, the user signs in as the test account themselves.
Reading, cheapest first
- REST, GET only. Exact, compact, no page to render.
Find the resource in the Oracle REST API reference first (web search restricted tonpm run fusion -- rest "/hcmRestApi/resources/11.13.18.05/<resource>?q=<filter>&fields=<a,b>&onlyData=true&limit=25"docs.oracle.com, thenfetch_oracle_page)..../<resource>/describereturns the attributes and child resources. Always narrow withq,fields,onlyData=trueandlimit: an unfiltered collection is thousands of lines. - Snapshot of a page, when no REST resource exposes the setting (most setup pages).
The output is the accessibility tree as text: roles, names, values. Scope it withnpm run fusion -- goto "<path on the pod>" npm run fusion -- snapshot --within "region=<section name>" --max 8000--within, a full Fusion page is long.
Large outputs go to disk with --out <file>, which only writes inside the engagement directory.
Changing something
Only in the test pod, and only a change the user owns.
A sequence of changes that forms a test scenario belongs to fusion-test instead: there, one
approval covers a whole written plan. What follows is for a single change outside any plan.
- State the exact change in the conversation: object, field, old value, new value, and how it will be verified. Wait for an explicit yes. One approval covers one change.
- Navigate and fill with
goto,click,fill,choose. None of these commits anything. - Commit with
click "<Save or Submit>" --write. Without--writethe tool refuses. - Verify by reading back, with a snapshot or REST, then retest under the target role. A retest
runs on subjects created for it, read before and after, never on an existing record whose
history is unknown (the controlled experiment of
hcm-debuggingstep 5). - Record it: a change note under
output/changes/in the engagement directory (seehcm-configuration), and the evidence line in the brief.fusion/audit.logthere holds the raw trail.
Every path walked with goto and click is evidence about the navigation. When it is not yet in
skills/fusion-reference/navigation.md, or differs from it, add or fix the row there with the date,
generic names only.
When a match is ambiguous the tool lists the candidates and stops, never guessing. Add --nth,
--exact or --within, do not retry blindly.
Hard rules
- Never production. Do not weaken the host guard, and never set
FUSION_ALLOW_UNRECOGNISED_HOSTwithout a written decision from the user. - Never a credential, never a sign in on the user's behalf.
- Every
--writefollows an explicit approval in the conversation, for that change. The only exception is a step of a test plan the user approved as a whole (fusion-test), and it covers that plan's steps only. - Client data read from the instance stays in the engagement directory and in the conversation,
nowhere else. That directory is outside this repository, named by
ENGAGEMENT_DIR. - Close the window at the end of the session:
npm run fusion -- close.
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.

