Skip to content

Checks, completion, and reports

What it is. Three shared contracts every Skill uses at closure:

Why it exists. “All tests pass” is the most expensive sentence an agent can say falsely. Kiyo makes every check, including the ones that did not run, an explicit record bound to the state that was actually checked.

Five check statuses

Section titled “Five check statuses”
Status Meaning
PASS The check actually ran, or the stated manual inspection was actually performed, and the observed evidence met its criterion within the inspected scope
FAIL The performed check did not meet its criterion, including an evidenced baseline failure
NOT_RUN The check was not performed; state why and whether it is still required. A newly written test has no passing result
NOT_APPLICABLE The criterion does not apply to this scoped task, with a concrete reason
BLOCKED An evidenced missing prerequisite, or a denied or unsafe target, prevents the check

A missing test environment is NOT_RUN or BLOCKED, never NOT_APPLICABLE. A mandatory check stays mandatory when it is inconvenient.

The nine-field check record

Section titled “The nine-field check record”
Field Content
Name Distinct check name and the behavior or acceptance criterion it addresses
Applicability Required, optional, or not applicable, with reason and source
Command/method Exact command and working context, or a clearly labeled planned method
Inspected scope Files, symbols, behavior, artifact state, environment, and target actually covered
Execution status Exactly one of the five statuses above
Observed result What actually happened; exit codes and counts only when observed
Evidence location An accessible reference or a concise in-conversation observation; never a fabricated log
Limitations Exclusions, partial output, uncertain attribution, untested environments
Baseline relation Existing failure, new regression, environment problem, unresolved, or not applicable, with the comparison source; otherwise Unknown

Results are bound to state

Section titled “Results are bound to state”

KIYO-VERIFY-002 keeps old results honest:

  1. Results bind to the inspected artifact, configuration, and environment, including uncommitted changes.
  2. After any further code, test, config, fixture, dependency, or environment change, affected results become NOT_RUN or BLOCKED until rerun. The earlier PASS stays historical.
  3. Never “all tests pass” while a baseline failure persists, a required check is unrun or blocked, or only a subset ran. The correct phrasing is: “the named focused check passed; the recorded baseline failure remains.”

The three evidence layers never upgrade each other: static inspection, behavioral evaluation, and live host tests. A DOCUMENTED_ONLY research note is not a behavioral PASS.

Four task statuses

Section titled “Four task statuses”
Task status Use when
DONE The named deliverable and all its mandatory completion criteria are satisfied with current evidence
PARTIALLY COMPLETE Useful scoped work was delivered, but requested work or mandatory evidence remains
BLOCKED A concrete missing prerequisite prevents required progress
DECISION REQUIRED An unresolved material business, scope, or authority choice blocks the next dependent action

Check statuses and task statuses are separate fields. A bounded review can be DONE with FAIL findings. A test run can be DONE with failing tests if running and reporting was the objective. Requirement readiness (READY_FOR_IMPLEMENTATION, DECISION_REQUIRED, INSUFFICIENT_EVIDENCE) is a third, separate field.

Definition of Done per workflow (summary)

Section titled “Definition of Done per workflow (summary)”
Workflow Done means Done does not mean
Implement Behavior and acceptance met, approvals still valid, required checks pass on the final state, reviews complete, Memory impact assessed and any mandatory sync done Production works; all consumers pass
Review Agreed scope inspected; findings, evidence, and limits delivered The application is bug-free or tests pass
Requirement Deliverable with separated facts, proposals, decisions, and unknowns Permission to implement
Test assess: analysis delivered · run: actual results reported · write: scoped tests delivered Tests passed (for write-only delivery)
Security Bounded assessment and required methods completed “Secure”, certified, or independently audited
Architecture Observed, approved, proposed, and unknown separated; findings delivered Decision adopted, implemented, or migrated
Memory show/check Records shown or compared; zero writes Whole-store currency
Memory sync/repair, Init Necessary authorized delta applied, or a real no-op Native activation or whole-project verification

The shared report

Section titled “The shared report”

Chat is the default output. Every report conveys eight fields; compact answers may combine them:

templates/reports/compact-task-report.md (artifact body)
- **Task/scope:** <requested result, mode, boundary and exclusions>
- **Actions/files changed:** <actual changes/inspection; proposed work distinguished>
- **Governance/risk rationale:** <brief scope/effect rationale, authority and uncertainties>
- **Verification/evidence:** <repeat the nine-field check record below as needed>
- **Residual issues:** <failures, untested scope, unknowns or none identified within scope>
- **Memory impact:** <one value; reason and actual applied/pending delta>
- **Status:** <one task status and its scope; readiness separately if relevant>
- **Next required action:** <concrete next step/precondition or none for completed scope>

A report file is written only within an authorized output scope. Reports exclude secrets, unnecessary personal data, raw logs, transcripts, and private reasoning. Self-review is labeled as self-review; a Markdown report is not a tamper-proof log.

Section titled “Related”