Install any skill in seconds. Free to start, no credit card required.
Get Started Free →hashintel/sdcpn-modelling
Elicit or revise an operational process model, maintain its recoverable workpiece, and construct a checked SDCPN when Petrinaut capabilities are available.
Passed 16 of 23 test cases with the skill, and 15 of 23 without it.
and 4 more in the full report
4 fixed12 passed either way4 still failing3 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
Our process net includes place capacity limits and transition delay parameters for order picking. I haven't formulated any specific decision or hypothesis I want the model to answer yet. Should we launch or propose an experiment run right now?
| Checked | Without skill | With skill |
|---|---|---|
| Refrains from proposing an experiment configuration when no decision goal is stated. | Passed | Passed |
| Explains that parameters in the net do not trigger experiment proposals by themselves. | Missed | Passed |
The instructions your model reads when the skill loads.
Interleave elicitation, workpiece settlement, supported construction and checks. These are recurring steps, not an interview phase followed by a construction phase. The current conversation may expose only one branch of that lifecycle. Do not claim that an unavailable transition occurred.
Interview in the person's operational vocabulary. Activate the elicitation skill and read references/profile.md before substantive questions or revision. Read templates/workpiece.md when creating or materially revising the shared workpiece. Construct only when the mounted capabilities actually permit construction in this conversation.
Establish enough purpose and context to select one focused next action: the intended question or decision, audience, boundary, horizon, accuracy need, and available time. Orientation need not settle every concern before elicitation begins.
For a new account, follow one concrete case and re-evaluate the active gap after each useful answer. For an existing account, first locate the disputed or changed material and its consequence for the objective. Use the elicitation skill's universal guidance and references/profile.md for detailed operations and coverage; do not turn their register order into question order.
Treat the workpiece as the recoverable operational account. Follow core's elicitation guidance for settlement cadence, evidence relations and locator lookup; templates/workpiece.md supplies the process-specific recording shape.
Settle the operational account with mutate_workpiece as meaning becomes clear.
Read references/pn-construction.md and references/checks.md before beginning. Use mounted Petrinaut tools for every net change and inspect the resulting definition rather than emitting free-form net JSON. If the required tools are absent, limit the result to the workpiece and construction-ready notes.
After each meaning-bearing answer, compare the supported account with the current net and take exactly one disposition before the next unrelated interview question. Changed: observe the current net, apply the bounded missing or changed fragment, run references/checks.md. Already represented: meaning the net already carries needs no mutation, judged from a current canonical read that stays reusable while the document does not change; reread when it is absent or the document changes. Blocked: the fragment lacks a load-bearing fact or the target cannot represent it; ask the smallest resolving question or record the exact missing fact or lost representation in the workpiece. Unrelated unknowns do not postpone supported construction.
Construction may infer a representation from established operational meaning; it may not invent operational facts without authorization. An explicit request to use sensible defaults, decide on the person's behalf, make up a suitable example, or equivalent authorization permits purpose-bounded assumptions without another interview round. Record construction inferences, labelled defaults, approximations and target losses in the workpiece. Labelling an unsupported operational default as an assumption does not authorize using it when that authorization is absent.
Saved scenarios and metrics are ordinary construction: build them when the person establishes a regime or a measure, not merely when an experiment is proposed. When the person states a decision the model should answer, read references/experiment-configuration.md as part of the same disposition and assess experiment readiness from what the person has stated and the current net's executability together. Propose the experiment when both are present, once per configuration; when only the net is missing something, construct it; when only a fact is missing, ask for it. Parameters or metrics in the net never trigger a proposal by themselves.
Apply references/checks.md whenever construction is prepared or attempted. Deliver the current workpiece in every branch. Deliver a net only when the mounted tool path has produced and checked one. State what the result can support, what remains open, what was assumed or simplified, and what the target or current tools could not represent.
An explicit stop opens no new topic. Emit the best current workpiece and any already-checked net with limitations visible.
test cases passed, vs 15 / 23 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.