01
Create a verifiable goal
Define the requested outcome, constraints, non-goals, affected surfaces, risks, acceptance checks, and evidence required for completion.
evidence-driven agent execution
CategoryAI & Codex Skills
Codex Krypton Skill
Codex Krypton Skill helps stop coding agents from shipping work that sounds plausible but has not been proven. Download the skill for planning goals, defining checks, executing controlled steps, collecting evidence, and refusing completion claims that are not supported by the repository or the running result.
Goal packagesExecution checksEvidenceCompletion gates
01
02
03
04
05
06
07
08
09
02
03
04
05
06
07
08
09
task: complete evidence-driven agent execution task
inspect:
- requirements and context
- existing standards
- failure and edge cases
verify: outputs + checks + handoffWhy a specialist workflow matters
A polished explanation can hide a broken build, an untested assumption, an incomplete requirement, or a result that was never inspected.
Krypton treats planning, execution, checks, and evidence as separate responsibilities connected by an explicit goal.
The skill helps Codex turn intent into a goal package, choose proof before implementation, preserve accepted evidence, and block unsupported completion claims.
What the downloadable skills can do
What Codex Krypton Skill can help accomplish
02
Plan executable checks
Choose commands, tests, inspections, artifact reviews, browser checks, or manual confirmations that can prove the goal.
03
Run controlled implementation
Follow the approved plan, limit scope, stop on contradictory evidence, and update the goal when reality invalidates an assumption.
04
Record and judge evidence
Capture relevant outputs and artifacts, distinguish passed checks from missing proof, and issue a clear completion or blocked result.
A repeatable working process
How the evidence-driven agent execution workflow moves from request to verified result
The steps keep context, implementation, and verification visible so the result can be reviewed and repeated.
workflow.statusREADY
Context → Plan → Work → Verify
01
Turn the request into a goal package
Record the outcome, scope, constraints, risks, acceptance criteria, and evidence that will count.
02
Design the proof before the change
Select checks that measure the actual result rather than only whether code compiles or a command exits successfully.
03
Execute against the goal
Make focused changes, run checks at useful milestones, and stop when failures or new information change the plan.
04
Close only with accepted evidence
Review outputs and artifacts, record what passed, identify missing proof, and avoid a completion claim when the gate is not met.
Useful across real projects
Where Codex Krypton Skill fit
The workflow adjusts to the project, audience, tools, and risk while preserving the same quality standard.
Implementation work
Connect a feature or fix to explicit acceptance checks and evidence from the changed system.
Quality gates
Create repeatable build, test, lint, policy, artifact, browser, or manual checks around important outcomes.
Agent workflows
Separate planning, execution, verification, and closeout responsibilities so plausible prose cannot substitute for proof.
Handoffs and audits
Preserve goals, decisions, evidence, failures, and remaining work for another person or agent to inspect.
Common requests
Tasks these skill can handle
Start with one defined outcome and provide the source material, constraints, and checks that matter.
01Define a verifiable goal
02Create an evidence plan
03Build execution checks
04Run a controlled implementation
05Capture test and artifact evidence
06Reject an unsupported completion claim
07Prepare an evidence report
08Resume a partially verified task
Who benefits most
Who Is This For?
These skill is designed for people who need dependable evidence-driven agent execution work with a visible process.
Developers using autonomous agents
Require proof that implementation actually satisfies the stated goal.
Engineering leads
Standardize acceptance gates and make agent completion reports easier to audit.
Quality and platform teams
Build reusable checks, evidence formats, and workflow constraints around repository changes.
Teams handing off long tasks
Preserve goal state and accepted evidence across sessions, people, and agents.
Good to know:
Krypton improves verification discipline, but evidence is only as good as the checks and environment selected. Human judgment remains necessary for incomplete specifications, subjective quality, security, safety, and high-impact releases.
Set up the workflow
Installation Guide
Install the complete skill folder and add the project-specific context before beginning.
Download and extract the Krypton skill
Keep SKILL.md with planning, execution, check design, evidence recording, and completion-gate guidance.
Install at project scope
Place the skill with the repository so goals and checks can reference its real commands, artifacts, and boundaries.
Define the evidence location
Choose where goal documents, accepted outputs, screenshots, logs, reports, and unresolved failures will be stored.
Set controlled-action rules
Document which commands are safe, which changes need approval, what cannot be automated, and who accepts the final evidence.
Before you download
Frequently Asked Questions
Practical answers about capabilities, limits, setup, and review.
What problem does Krypton solve?+
It reduces plausible but unverified work by requiring a goal, suitable checks, captured evidence, and a completion gate.
Does it replace tests?+
No. Tests are one form of evidence. Some goals also require artifact inspection, browser behavior, operational signals, or human acceptance.
Can it work with an existing plan?+
Yes. It can convert an approved plan into explicit checks and evidence requirements before execution.
What happens when a check fails?+
The task remains incomplete. Codex records the failure, investigates within scope, and updates the plan or reports the blocker.
Can Krypton prove a system is correct?+
No. It can show that defined checks passed in a stated environment. It cannot prove properties the checks did not cover.
Make the work repeatable
Give Codex an execution workflow where goals are explicit, checks measure the result, and completion requires evidence.
Clear context. Purposeful work. Relevant checks. A result others can understand.