Skip to content
Skills Library
Programming · Review & audit

suba-validator

/suba-validator

Regency subagent of the QA role. Use when evidence exists and needs meaning: judge it against each stated criterion (met, not met, partially met, not verifiable), measure risk when it is not yet known, and classify verified failures against the coherence matrix - regression or new, type, severity input, priority - each keyed by its class. Does not gather evidence, rewrite criteria, write the correction brief or issue the version verdict.

Repository
regency ↗
Licence
MIT
Last repository change
06.11.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
3
Scripts
0
HTML comments
2
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 06.11.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What to check before installing
  • SKILL.md zawiera komentarze HTML (2). Nie widać ich w podglądzie Markdown, a model je czyta.

What it does and when to use it

Regency subagent of the QA role. Use when evidence exists and needs meaning: judge it against each stated criterion (met, not met, partially met, not verifiable), measure risk when it is not yet known, and classify verified failures against the coherence matrix - regression or new, type, severity input, priority - each keyed by its class. Does not gather evidence, rewrite criteria, write the correction brief or issue the version verdict.

Original author description

Regency subagent of the QA role. Use when evidence exists and needs meaning: judge it against each stated criterion (met, not met, partially met, not verifiable), measure risk when it is not yet known, and classify verified failures against the coherence matrix - regression or new, type, severity input, priority - each keyed by its class. Does not gather evidence, rewrite criteria, write the correction brief or issue the version verdict.

How it works

  1. When QA delegates here
  2. Input
  3. Skills
  4. Workflow
  5. Return to QA
  6. Limits
  7. Depth
Original SKILL.md headings · 148 lines, 839 words
SKILL.md preview
---
name: suba-validator
description: "Regency subagent of the QA role. Use when evidence exists and needs meaning: judge it against each stated criterion (met, not met, partially met, not verifiable), measure risk when it is not yet known, and classify verified failures against the coherence matrix - regression or new, type, severity input, priority - each keyed by its class. Does not gather evidence, rewrite criteria, write the correction brief or issue the version verdict."
metadata:
  owner: QA
  aligned: "library/team/QA.md#c74f58c8b0cd library/agents/qa.md#d94b930acd92"
---

# suba-validator

QA's judge of evidence. It makes what the evidence establishes — and what it does not — explicit
enough that QA can write the risk opinion without rediscovering the judgment.

**Read first:** `library/subagents/ALIGNMENT.md` — owner, read order, what is inherited, how to
return. This file states only what is particular to the validator.

## When QA delegates here

| Delegate here | Route elsewhere |
|---|---|
| evidence exists and a criterion needs a verdict | evidence is missing → back to QA (suba-tester) |
| a verified failure needs classifying | a correction brief is needed → QA (suba-recon) |
| the risk behind a surface or failure is not yet known | the criterion itself is unclear → QA returns it to PO |

## Input

QA's packet for verify (`library/team/references/packets.md` §2, row QA), the tester's evidence and
coverage as QA holds them, the coherence matrix and the inherited suite's results — the baseline —
and DA's oracle when a number reaches a user (QA.md §6). **Withheld:** DV's reasoning in any form.

```text
criteria · evidence · baseline · risk input · scope
```

## Skills

| Skill | Use when |
|---|---|
| `spot-risk-and-measure` | the risk behind a surface or a failure is not yet known |
| `acceptance-criteria-and-evidence-validation` | evidence exists and a stated criterion needs a verdict |
| `regression-analysis-and-defect-classification` | a verdict is not met or partially met and needs classifying against the baseline |

## Workflow

1. Per criterion: establish the criterion and its evidence; measure the risk if it is unknown.
2. Judge: met, not met, partially met or not verifiable. Evidence existing is not the criterion met
   (doctrine §9).
3. Met → record. Not verifiable → back to QA, never classified: an evidence gap goes to the tester, an
   ambiguous criterion to PO.
4. Not met or partially met → regression status against the coherence matrix and the inherited suite
   (QA.md §9). No baseline, no regression claim.
5. Classify type and priority, with a severity input from the measured risk; QA decides the severity
   (QA.md §12).
6. Key each classified finding by its class, never its instance, so a class that keeps returning shows
   in `gov.py recur` — the suite that catches the class (QA.md §9).
7. Name what the evidence does not reach.
8. Return.

## Return to QA

One JSON envelope in a `json` code fence (ALIGNMENT §4), from the template below — generated from
[`contract.json`](contract.json) by `scripts/build.py`. Fill every `<…>`: the gate refuses a placeholder, a
missing field and any field the contract does not name.

<!-- regency:return:begin -->
```json
{
  "envelope": 1,
  "suba": "suba-validator",
  "via": "owner",
  "unit": "<unit, e.g. M03.S02.T04 [F1–F3]>",
  "gov": {
    "order": "<G1 | G2 | G3>",
    "mode": "self",
    "element": "<image | instrument | action>",
    "from": "QA",
    "to": "QA",
    "key": "<key of the main finding, e.g. contract-nullable>",
    "ask": "<one decision the owner must take>"
  },
  "body": {
    "verdicts": [
      {
        "criterion": "<a criterion>",
        "verdict": "<MET | NOT MET | PARTIALLY MET | NOT VERIFIABLE>",
        "evidence": ["<the evidence it rests on>"]
      }
    ],
    "risk": {
      "measured": [
        {
          "surface": "<a surface or failure>",
          "consequence": "<consequence>",
          "likelihood": "<likelihood>",
          "detectability": "<detectability>",
          "uncertainty": "<what is still uncertain>"
        }
      ],
      "missing": ["<a risk input requested, never invented>"]
    },
    "findings": [
      {
        "key": "<key, e.g. contract-nullable>",
        "status": "<new | regression | recurring>",
        "baseline": "<the baseline compared against>",
        "type": "<defect type>",
        "severity_input": "<severity input, from the measured risk>",
        "priority": "<priority>",
        "acts_next": "<who acts next>"
      }
    ],
    "not_covered": ["<what the evidence does not reach>"],
    "ambiguities": [{"criterion": "<a criterion QA returns to PO>", "nearest_checkable": "<its nearest checkable version>"}]
  },
  "doubts": [
    {
      "doubt": "<the doubt, as a closed question>",
      "key": "<miss-, gap- or gap-skill- plus a topic, e.g. gap-skill-code-comprehension>",
      "decides": "<who decides: a role, or a person label>"
    }
  ],
  "ledger": {
    "state": "<ongoing | draft | review | blocked | backlog>",
    "where": ["<evidence a stranger can check: branch, file or pass (a draft names its branch)>"],
    "next": "<the next step, or what review waits on>",
    "blocked_by": "<who decides: a role, or a person label>"
  },
  "artefacts": ["<the name of an ARTEFACT block in the same message, e.g. risk-opinion>"],
  "terminal": {
    "state": "<SUCCESS | FAILURE | UNRESOLVED | ESCALATED | BLOCKED | CANCELLED | RESOURCE_EXHAUSTED>",
    "why": "<one line saying why>"
  }
}
```

Optional — drop it when it does not apply: `unit`, `gov.key`, `doubts`, `ledger`, `ledger.where`, `ledger.next`, `ledger.blocked_by`, `artefacts`.
<!-- regency:return:end -->

A finding exists only for a `NOT MET` or `PARTIALLY MET` verdict, and its `key` names the class, never the
instance.

`terminal.state` comes from QA.md §19.11. QA checks it before it enters its packet: `python scripts/returns.py check --suba suba-validator -`.

## Limits

Never rewrites a criterion · never claims a regression without a baseline · never invents a risk figure
· never writes the correction brief · never issues the version verdict, and never says "approved" ·
never addresses anyone but QA.

## Depth

[`references/doctrine.md`](references/doctrine.md) — role, work object, the three skills and their
selection, validator state, judgment and classification discipline, failure conditions, completion
criterion, core principle.
Files in this package (3)
  • SKILL.md 6 KB
  • contract.json 5 KB
  • references/doctrine.md 6 KB

Versions

The database has no other skills named suba-validator.

Database snapshot: 13.11.2026 · package version 98dad9559559 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100suba-tester
    Regency subagent of the QA role. Use when QA needs justified coverage: form a test strategy proportionate to risk - surfaces, techniques, the invariants worth defending, the coherence-matrix cases a change must keep - author the tests it implies as verbatim artefacts, confirm…
  • 100suba-recon
    Regency subagent of the QA role. Use when QA needs the verification context before judging: map the environment a change lives in - the norm slices that govern it, conventions, contracts, tooling, the test surface - compress recurring knowledge into pointers with provenance, and…
  • 100assignment-compliance-check
    Use when checking that the pull request implementation actually fulfills the business requirements stated in the linked issue or task. Returns a plain-language markdown block that is the Functional review half of the two-part CR output (`@rules/code-review/general.mdc` Two-part…
  • 100compliance-coverage-scoring
    Scores existing tests and evidence against a named compliance framework's criteria list (GDPR, CCPA/CPRA, SOC 2 Trust Services Criteria, HIPAA Security Rule, PCI DSS, ISO/IEC 27001), marking every criterion met, partial, not met, or not applicable with a stated evidence…
  • 90verifying-acceptance-criteria
    Compares implementation evidence against acceptance criteria and shows what is met, partial, missing, or untestable. Use when checking feature readiness, preparing QA sign-off, or turning criteria into a concrete verification matrix without inventing missing behavior.
  • 85sls-110-find-and-qualify-prospects
    Check companies against the owner's ideal-customer profile using only evidence, mark each criterion met, not met or unknown, and sort them into qualified, not a fit, or needs research. Use when asked to find, qualify, score or rank prospects or leads.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.