Turn process evidence into a measurable current-state diagnosis, safer future-state options, and a controlled pilot with gates, rollback, and verification.
Most process reviews jump from a slow queue to an automation idea. That skips the hard part: establishing what entered the process, where work waited, which variants behave differently, what caused rework, and which controls cannot be removed casually.
Process Improvement builds that decision record. It defines one trigger-to-completion process, separates observed data from reported claims, maps exception and rework paths, reconciles the work-item population, and creates a baseline with explicit windows and denominators. Material issues stay linked to sources, metrics, competing explanations, and falsification tests.
Use it to diagnose invoice approval, customer onboarding, access review, service operations, internal requests, and other recurring professional workflows. The output compares future-state options, shows control and downstream effects, and turns the selected change into a bounded pilot with guardrails, stop conditions, rollback, owners, and verification. Proposed, approved, implemented, and verified changes remain separate.
Version 1.0.0 was reviewed on 2026-07-17 and evaluated through Codex CLI with gpt-5.6-sol in 12 fresh read-only contexts. The candidate scored 100 across three cases, compared with 71.25 for the same agent without the skill and 83.33 for the exact free upstream. The cases covered invoice approval under pressure to remove controls, B2B onboarding with hidden process variants, and a canonical machine-checkable record with blocker precedence. This evaluation measures the supplied cases. It does not prove a real bottleneck, causal improvement, regulatory compliance, cost savings, or a safe live change. The reviewed package SHA-256 is 452079fbed3dec56f04d89d9170c55717006486fb35282a5a7f9c10e566e386e.
The package includes a read-only checker for an optional canonical JSON process record. Passing it confirms structure, references, step links, evidence states, gates, rollout state, and verdict precedence. It cannot inspect source systems, verify source truth, measure the real process, approve a change, or prove that a pilot worked. Maintenance and install support are handled by the platform-operated publisher for this listing.
Input
Review our last quarter of invoice flow. We had 684 invoices, long approval waits, incomplete requests, wrong routing, and twice-weekly payment runs. The manager wants no approval below $5,000 and an annual savings claim, but the policy authority and labor rate are missing.
Output
A reconciled source and work-item record; current-state map with amount variants, return loops, approval queues, and payment batching; baseline and evidence gaps; tested cause hypotheses; control-preserving options; a bounded pilot with metrics, guardrails, stop conditions, rollback, owners, and a blocked or review-ready verdict.
Input
Improve contract-to-first-sync onboarding. CRM and provisioning populations do not reconcile, SSO takes much longer, setup timestamps are incomplete, and teams are debating an input gate, earlier security review, or automatic field mapping.
Provide one recurring process from an observable trigger to an observable completion event. Include source records or samples, observation dates, work-item counts, timing, variants, exception and rework paths, owners, systems, controls, service targets, current pain, and the decision and permission boundary. Ask for diagnosis, option comparison, a pilot plan, or verification of an implemented change. The skill keeps missing evidence and authority visible instead of inventing savings or removing controls by assumption.
One recurring trigger-to-completion process, work-item unit, included variants and teams, target outcome, observation window, and the decision this work must support.
Timestamped exports, logs, samples, policies, SOPs, interviews, queue and rework records, service targets, and known limitations with stable locators or revisions.
Applicable financial, legal, safety, privacy, security, accessibility, segregation, customer, system, staffing, and change-window requirements.
Process owner, change and control approvers, pilot owner, affected participants, and whether the agent may only inspect and plan or may also implement a bounded change.
SKILL.md; evidence, mapping, diagnosis, option, pilot, and verification method; canonical Markdown and JSON output contract; read-only process-record checker; Agent Skills interface metadata
No reviews yet.
Output
Separate standard and SSO maps; population reconciliation; segmented lead-time and rework baseline; source-state and ownership gaps; comparison of input, security, and mapping options; mapping and security guardrails; a small pilot design; and exact gates before rollout.
Input
Create the canonical JSON record for an access-review process with observed step timing, one queue hypothesis, two reversible options, an identified but unapproved control owner, an open approval gate, and no after-period result.
Output
A schema-bound record with reciprocal step links, resolvable sources and authorities, observed baseline and missing after measurement, issue and option references, an open critical gate, a proposed rollout, and a blocked verdict that the bundled checker can validate.
Trebuh
Updated Jul 13, 2026
Automated scan passed. It found 1 finding; highest severity info.
Scanned Jul 17, 2026.
Automated scanning reduces risk but cannot guarantee a package is safe for every environment.
Creator
Ffiralo30
