The Enso Method
Turn a project idea into clear requirements, practical build phases, and evidence that the result works. A self-contained, MIT-licensed SkillDB adaptation of EnsoDynamics' method, with proposed answers and stakeholder corrections kept alongside the work.
Use this workflow when a project needs to stay understandable from the first request through implementation and review. The central habit is simple: write down what the product should do, record uncertain answers honestly, and keep those records current as the work progresses. A new session should be able to continue from the project files without reconstructing a long conversation.
skilldb get enso-method-skills/the-enso-methodFull skill: 121 linesThe Enso Method
Use this workflow when a project needs to stay understandable from the first request through implementation and review. The central habit is simple: write down what the product should do, record uncertain answers honestly, and keep those records current as the work progresses. A new session should be able to continue from the project files without reconstructing a long conversation.
This is a self-contained SkillDB adaptation of the method published by EnsoDynamics. It is usable as written in an agent's normal workspace; it does not require the upstream plugin, helper scripts, or an external installation. SkillDB's examples and organization adapt the method for this catalog. The upstream copyright and MIT permission notice are included below.
Start with the project's actual state
Read the user's current request, project instructions, durable memory, relevant requirements, and the current repository status. Separate the intended product from what files, tests, and deployed behavior actually show. Do not restart a completed stage merely because the current conversation is new. Do not overwrite existing document layouts to impose this example's filenames.
Use this routing table to choose the smallest useful next step:
| What you find | Next action | Concrete output |
|---|---|---|
| Ideas, conversations, or conflicting requests | Gather decisions and write requirements | A scoped requirements document with acceptance criteria |
| An important unanswered business question | Research and propose an answer | An assumption with evidence, confidence, and an owner |
| Clear outcomes but an uncertain implementation | Design and check readiness | A design mapped to requirements and identified gaps |
| Work too large to verify in one slice | Split the initiative into phases | Dependency order, criterion ownership, and completion evidence |
| A ready phase within the user's authorization | Implement and verify that phase | Changed artifacts, test results, and a checkpoint |
| A reported defect with an unclear cause | Investigate before choosing a fix | Reproduction, evidence, and a bounded proposed correction |
| A stakeholder correction | Reconcile the affected source documents | Updated requirements, assumptions, design, and remaining phases |
| A build reported as finished | Review delivery and test evidence | Criterion-level findings, exceptions, and remaining work |
The other seven skills in this pack expand those procedures: Requirements and PRD; Assumptions and Stakeholder Feedback; Technical Design and Readiness; Initiative and Phase Planning; Implementation and Recovery; Evidence-Based Review and Testing; and Stakeholder Review Packet. Each can also be used independently.
Keep a small set of durable records
Use existing project paths when available. For a new project, a useful minimum is:
requirements.md Intended behavior, scope, criteria, source decisions
assumptions.md Proposed answers, evidence, confidence, corrections
technical-design.md Components, interfaces, failure behavior, verification
phases/ Ordered slices and their current state
evidence/ Reproductions, test results, and review findings
handoff.md Current checkpoint and the next bounded action
Give consequential criteria stable identifiers. A requirement such as “refunds should work” cannot distinguish a correct result from a misleading success screen. Write the actor, input, preconditions, observable result, and rejected case. Keep implementation choices in the design unless they are explicit product constraints. Link decisions to their source and date, and link evidence to the exact version it exercised.
For example, Project Returns uses synthetic orders to develop a support refund workflow. R-01 permits an authorized support representative to approve a full refund to the original payment for an eligible order. R-02 prevents a retry from creating a second refund and rejects reuse of an operation ID with a different request. R-03 returns an appropriate safe result for an ineligible or unauthorized request without a provider call or a committed refund. R-04 preserves an uncertain provider outcome as pending until authoritative reconciliation. Real payments are outside this example's authorization.
Make uncertainty useful
For each material unanswered question, record a proposed answer when the evidence supports one. Include the source, what it actually establishes, contrary evidence, confidence with a rationale, the responsible decision maker, and what changes if the answer is corrected. Confidence describes the strength of a proposal; it is neither approval nor a measured probability unless a real measurement supports it.
A-01: What refund eligibility window applies?
Proposal: Through 30 elapsed 24-hour days after the recorded payment, inclusive, using the recorded request time in UTC.
Evidence: Current policy draft; sample historical cases are incomplete.
Confidence: Medium; the draft does not resolve older orders.
Status: Provisional; policy owner has not confirmed.
Affected: R-01, R-03; eligibility boundary checks; phase PH-01.
If corrected: Change the rule and boundaries before enabling real refunds.
Permitted work: Synthetic rule examples and interface preparation.
Continue useful work that is authorized and does not depend on an unresolved consequential decision. A reversible draft can use a clearly marked assumption. Spending money, changing access, contacting people, or publishing is governed by the user's actual authorization and project controls. An unanswered question or elapsed time never grants that authority. If no defensible answer exists, record the question and its impact instead of inventing a confident proposal.
Design, phase, and build against the same criteria
Translate requirements into an implementation that explains data ownership, interfaces, permissions, error handling, migration, and how each criterion will be demonstrated. For Project Returns, the design must say where an operation ID is persisted, how a repeated request finds the earlier outcome, and how an uncertain response is reconciled. Writing “make refunds idempotent” leaves the hard behavior unspecified.
Split work by demonstrable outcomes. PH-01 might implement eligibility checks against synthetic records. PH-02 can add a test-mode provider adapter and retry handling once the eligibility contract is stable. PH-03 can verify authorized end-to-end behavior and operational recovery. Assign every criterion an implementation owner and an evidence location. A stubbed provider can support development; it cannot establish what the real provider does.
For each phase, read its current inputs, implement the bounded change, run relevant checks, and record the result. Mark a phase complete only when its exit conditions have supporting evidence or the user explicitly accepts a documented exception. Preserve failing results and partial runs. A green command exit is useful evidence for that command, not proof of the entire product.
Reconcile corrections before they disappear
Suppose the policy owner corrects A-01 to 45 elapsed 24-hour days, still inclusive, and specifies an effective date. Record who supplied the correction and its scope, revise the relevant criteria, update the design's rule, and identify boundary tests and phases that now require adjustment. Already completed evidence for the 30-day rule remains historical. Label it superseded for the affected behavior. Do not leave the requirement at 30 days while silently changing code to 45.
Distinguish a correction to agreed behavior from a new feature request. A request for partial refunds introduces new amount, balance, and interface decisions; it is not automatically covered by the original full-refund criteria. Capture it as additional scope, identify its owner and dependencies, and proceed only within the user's authorized work. Unrelated phases can continue when their inputs remain valid.
Review outcomes and evidence separately
Perform a delivery review to determine whether each required behavior exists. Then inspect test quality: inputs, assertions, negative paths, side effects, environment, and whether the checks could detect the failure that matters. A duplicate-refund test should assert one provider effect after repeated requests, not merely two HTTP success responses. An ineligible-order test should assert the unchanged payment as well as the displayed explanation.
Record a compact matrix with criterion, implementation location, evidence, result, limitations, and next action. For R-04, a mocked timeout demonstrates local pending-state behavior; an approved sandbox exercise may be needed to demonstrate reconciliation against the provider's actual records. Keep those claims distinct. If evidence is unavailable, name the exact gap instead of calling the whole phase verified.
At the close of a multi-phase requirements document, trace every required variant from its real entry point. If completion work changes the implementation, revisit whole-document delivery against the current revision before finalizing the test audit. Preserve a known gap with its owner and next action until the evidence supports closure.
Leave a recoverable checkpoint
Before a handoff or session end, record the source revisions read, completed changes, checks and their outcomes, current assumptions, unresolved findings, and one next action. For external operations, retain a nonsecret receipt or operation identifier and distinguish completed, failed, and unknown outcomes. Reconcile unknown outcomes before retrying them.
Measure expensive waits once. Run independent read-only checks concurrently when they do not contend for fixtures or shared outputs. Reuse valid evidence until relevant inputs change. Do not weaken a test, skip a required gate, create an automation, or repeat a production action merely to shorten elapsed time.
The workflow has succeeded when another person or agent can identify what is required, what is provisional, what has actually been demonstrated, and what to do next from the project records alone. A long document without that traceability fails the same check as an undocumented build.
Attribution and license
Adapted from EnsoDynamics/enso-method, revision 23a908030bf6b4e228ba90f1937af87aed470e0b: using-the-enso-method and wall-clock-awareness. This catalog adaptation is distributed under the MIT License below.
MIT License
Copyright (c) 2026 Doug Kerwin
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OF OR OTHER DEALINGS IN THE SOFTWARE.
Install this skill directly: skilldb add enso-method-skills
Related Skills
Enso Method Research Assumptions and Reconcile Feedback
Use the Enso Method to record proposed business decisions with evidence and confidence, then propagate authorized stakeholder corrections to requirements and dependent work.
Enso Method Evidence-Based Review and Testing
Review deliverables against intent, audit delivery and tests by acceptance criterion, and verify repaired evidence at the boundary where the outcome is observed.
Enso Method Implementation and Recovery
Implement a bounded requirements phase, resolve discoveries in its source documents, and recover interrupted work without replaying uncertain effects.
Enso Method Initiative and Phase Planning
Turn an initiative into bounded requirements documents and implementation phases with stable acceptance criteria, dependency contracts, and reviewable handoffs.
Enso Method Requirements Intake and Testable PRDs
Use the Enso Method to turn source material and stakeholder decisions into an authoritative product requirements document with verifiable acceptance criteria.
Enso Method Stakeholder Review Packet
Prepare an approachable project review with selected requirements, every open business assumption, clear proposed answers, and traceable corrections. A self-contained, MIT-licensed SkillDB adaptation of EnsoDynamics' stakeholder review workflow.