القائمة
Use when running, reviewing, or preparing computational experiments so inputs, outputs, manifests, failure states, and generated artifacts remain explicit and reproducible.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Audit the structural and logical integrity of a mathematics paper — quantifier scope, domain-of-definition guards, proof/computation honesty, theorem hierarchy, contribution-list discipline (mapping every claimed contribution to its formal result, prior baseline, concrete gain, and independent significance), stale claims, notation accuracy at introduction, standing assumptions, and undefined relation symbols or coined terms in summary prose — without touching prose style or rhetorical prose inflation (structural inflation of contribution lists and result hierarchy is in scope).
Audit a LaTeX mathematics document for notation drift — symbols used under different names across sections, orphaned macros, undeclared aliases (including cross-language pairs where a term defined in one language is later used in another), symbols or relation signs used with no definition site anywhere, symbol reuse across scopes, and first-use-after-gap without back-reference — without evaluating mathematical correctness or prose style.
Review or rewrite mathematical prose so an edit, paraphrase, terminology migration, or explanatory description preserves the exact mathematical meaning fixed by its semantic source — referent identity, operation versus result, domain/codomain and representation level, quantifier and equality mode, role and provenance, notation role, and classified-before-replacement terminology migration with concept-cluster synchronization — without auditing claim-versus-proof strength (use math-claim-integrity), notation bookkeeping (use math-notation-consistency), Japanese style or translation events (use wabun-math-style), or audience prerequisites and ordering (use exposition-flow).
Detect and correct Japanese-language anti-patterns in mathematical writing — epistemic hedges weakening proved claims, incorrect verb tense, passive/active confusion in proofs, ambiguous particles, unjustified 明らか, decorative connectives, redundant meta-discourse, double negation, stacked の, generic/general conflation, borrowed non-mathematical vocabulary, vague category names for explicit formulas, full-width semicolons or untranslated English, canonical-terminology drift, and translation-level concept conflation — without evaluating mathematical correctness or structural hierarchy.
Turn repository-grounded source material into a self-contained, audience-admissible document or presentation by modelling the intended audience, extracting information units with explicit discourse roles, and enforcing dependency-ordered exposition in which every load-bearing statement follows the prerequisites needed to interpret it — without factual verification or citation checking (use sot-integrity), prose polishing (use deslop-prose), mathematical-correctness review (use the mathematical-writing skills), or typography and slide layout.
Workspace-level Git cleanup — discover repositories under the current directory and prune merged branches, merged linked worktrees, stale remote-tracking refs, and stale worktree metadata.
| name | computational-reproducibility |
| description | Use when running, reviewing, or preparing computational experiments so inputs, outputs, manifests, failure states, and generated artifacts remain explicit and reproducible. |
Use this skill to keep computational experiments reproducible before expensive execution or change submission.
The goal is experimental discipline: fail early on invalid setup, keep source and generated artifacts separate, record enough identity to rerun work, and report scientific outcomes without ambiguous status labels.
Treat each computation as a scientific claim, not merely working code. If reproducing the result faithfully would require equations laced with ad-hoc case splits, divide-by-zero guards, domain clipping, tuned regularization constants, or fallbacks, it has drifted from science toward engineering expedience: such devices belong in a scientific account only as part of the declared method — chosen deliberately, recorded, and visible — never inserted silently to make a degenerate case disappear. When in doubt, hard-fail and surface the degeneracy rather than patch past it.
Use this skill when:
Do not use this skill for:
Missing inputs, layout mismatch, dependency absence, unavailable compute, or unavailable API access must hard-fail before execution.
Fallbacks are allowed only when all of these are true:
Do not silently substitute sample data, cached outputs, smaller models, different backends, local paths, or alternate datasets.
The same rule governs fallbacks inside a computation, not only at its inputs. Replacing a degenerate intermediate or estimator with a default hard-fails by default; take the fallback path only when the same four conditions hold (explicit opt-in + recorded + visible in outputs + distinguishable in review). Examples across domains:
x / (denom + 1e-8)) instead of confronting the degeneracylog/sqrt/arccos, or clipping probabilities into (0, 1))This forbids silent divide-by-zero avoidance and domain clipping specifically: they may never be applied without the explicit opt-in above. Declared, recorded regularization or clipping that is the method (e.g. ridge regression, gradient clipping) is fine; the prohibition targets ad-hoc guards inserted to make a degenerate intermediate disappear.
Judge such a fallback by impact if triggered, not only by whether it fires on current data. A guard that never fires now can still be scientifically catastrophic if it would — for example a substituted scale that silently yields an order-of-magnitude-wrong result without signalling an error. A fallback that fires on a non-trivial fraction of inputs is itself a signal that the primary path's definition may be wrong; investigate the spec before accepting the fallback.
Keep source files and generated artifacts distinct.
Source material may belong in git:
Generated artifacts stay outside git by default:
Commit generated artifacts only when the repository explicitly treats them as source, fixtures, or reviewed deliverables.
Create or update the run manifest before expensive execution starts.
Record the applicable identity fields:
If a field does not apply, record that it is not applicable rather than omitting it when omission would create ambiguity.
Expensive execution starts only after required inputs, dependencies, identities, and compute access are validated.
Use stable source/output separation and stable run identity. Concrete directory names are repository policy.
Before launching long-running work, validate:
If preflight fails, stop before execution. Do not repair by guessing paths, creating ad-hoc fallback inputs, or switching execution backends silently.
Manifest and artifact layout must make each run auditable without prescribing repository-specific paths.
The manifest must identify source revision, command, configuration, input identity, output identity, seed status, execution backend, applicable model/service identity, environment identity, and fallback status.
Use stable source/output separation and a stable run identity. The exact directory names are repository policy; this skill only requires that:
Use distinct status terms.
not-run: execution did not start.skipped: execution was intentionally bypassed for a recorded reason.failed: execution started and did not complete successfully.empty-result: execution completed, but produced no result where a result may normally exist.zero-result: execution completed and the measured value is legitimately zero.Do not use empty, failed, or not-run interchangeably. If a status is uncertain, mark it as unknown and explain what evidence is missing.
When running an experiment:
When modifying an existing workflow, preserve stricter behaviour. Do not add silent fallback, implicit path discovery, runtime substitution for degenerate intermediates, or source/artifact conflation for convenience.
Before submitting changes involving computational experiments or generated outputs, check:
not-run, skipped, failed, empty-result, and zero-result.or followed by a constant, if x <= 0: x = ..., a small constant added to a denominator (+ 1e-8, + eps), clip/clamp into a valid domain, a default or initial value supplied to a reduce/fold/max/min, nan_to_num and equivalents, NaN-skipping reducers such as nanmean/nanmax/nansum (silent drop), and try/catch blocks returning a default.Stop and ask for a repository-specific decision when: