Skip to content

opensiro/vsm-harness-skills / skills/assess-vsm-harness

assess-vsm-harness

Verified

Produce a standalone evidence-backed VSM assessment of one autonomous AI agent harness at a pinned repository revision, including first-party ownership modes.

references/negative-state-calibration.md

View on GitHub
# Negative-state calibration — bounded absence and unresolved evidence

**Methodology version:** 0.3.8

**Status:** adjudication examples / calibration guide; not a new VSM definition, classification symbol, or assessment field. Profile 0.2.4 and existing Methodology thresholds remain authoritative.

## Procedure

1. Pin upstream source revision and declare the system-in-focus, recursion, supported standard-distribution modes, credited first-party surfaces, and explicitly excluded adjacent surfaces.
2. **Test positive ownership before negatives.** For each function map the disturbance, decisive organizational right, owner, and return/closure. Check supported optional modes and, for S3/S4/S5, legitimate parent-governed paths before concluding absence.
3. Trace **plausible first-party paths checked** that could materially close the function in the declared shipped mode. Use primary code wiring, supported configuration, actual runtime entry points, tests, and operating documentation. A component label is not a function witness.
4. Choose `—` only with a defensible **bounded** no-material-first-party-path conclusion and the existing `Absence scope`: (a) surfaces inspected, (b) plausible first-party paths checked, (c) why no material path remains. Exhausting *every imaginable extension* or proving impossibility is not required.
5. Choose `?` when reviewed primary evidence is insufficient for **both** a positive result and that bounded negative conclusion—for example, a specific plausible *shipped* path is unresolved, the supported mode cannot be examined, or material boundary coverage is incomplete. State the actual unresolved question in the existing Evidence gaps section where identifiable.
6. Never substitute historical corpus frequency for source evidence. **The absence of a positive witness alone does not establish `—`**: the negative conclusion still requires the existing boundary-relative `Absence scope`. Likewise, a hypothetical future plugin or feature alone does not require `?`.

A retired, ignored, test-only, dogfood-only, or third-party mechanism does not establish a positive path at the stated boundary. Before concluding `—`, examine whether there is a separate *active shipped* path; if that specific path cannot be resolved, `?` may be justified. Ordinary permission or configuration control alone does not establish S5, just as routine QA does not automatically establish S3*.

## Historical comparison anchors — not automatic gold labels

These published Index documents are pinned to the immutable [Index tree `012cfeec80d2c98bec90942d2edb66e5a6b71774`](https://github.com/opensiro/vsm-harness-index/tree/012cfeec80d2c98bec90942d2edb66e5a6b71774); retain their embedded upstream `review_ref` and assessed modes when comparing. They illustrate distinct *types* of evidence, not a rule to copy old decisions.

| Assessment | Published result | What distinguishes its evidence |
| --- | --- | --- |
| [ndx](https://github.com/opensiro/vsm-harness-index/blob/012cfeec80d2c98bec90942d2edb66e5a6b71774/assessments/ndx.md) | S5 = `—` | Inspects system instructions, permission policy, session/configuration, skill resources and collaboration controls; documents absent identity-level decisive-return path. |
| [Maki](https://github.com/opensiro/vsm-harness-index/blob/012cfeec80d2c98bec90942d2edb66e5a6b71774/assessments/maki.md) | S4 = `—` | Examines memory, provider/skill selection, planning and task modes; no supported outside-and-then capability adaptation closure remains. |
| [SharpCoder](https://github.com/opensiro/vsm-harness-index/blob/012cfeec80d2c98bec90942d2edb66e5a6b71774/assessments/sharpcoder.md) | S3* = `—` | Generic child review possible, but explicit absence scope finds no independent direct-evidence audit judgment with returned corrective control. |
| [Plank](https://github.com/opensiro/vsm-harness-index/blob/012cfeec80d2c98bec90942d2edb66e5a6b71774/assessments/plank.md) | S3* = `?` | Reviewer/hook surfaces leave an unresolved in-boundary independent corrective-audit question. |
| [GhostyCode](https://github.com/opensiro/vsm-harness-index/blob/012cfeec80d2c98bec90942d2edb66e5a6b71774/assessments/ghostycode.md) | S3* = `?` | First-party verifier/reviewer and raw diff/test surfaces are present, but real independent corrective ownership remains uncertain. |
| [pi-rs](https://github.com/opensiro/vsm-harness-index/blob/012cfeec80d2c98bec90942d2edb66e5a6b71774/assessments/pi-rs.md) | S4 = `A` | Positive evidence must be adjudicated on its own owner and prospective closure instead of being forced into a negative-only workflow. |

A historical example might itself be corrected after independent same-ref investigation. The calibration examples do not replace reviewing the pinned upstream evidence.

## Reviewer questions

- For `?`: what **concrete supported** plausible path or coverage gap prevents a defensible `—`? If none is identified, re-examine whether the written no-material-path conclusion is already supported.
- For `—`: are all three absence-scope elements substantive and tied to the *declared standard-distribution boundary*? If not, inspect further or retain `?`.
- For both: did the reviewer test the positive and parent-governed alternatives first, distinguish supporting enforcement from decision ownership, and avoid borrowing adjacent/external actors?

## Compatibility and correction

No field, state definition, positive threshold, burden of negative evidence, synthesis/ranking operation, or mandatory new artifact section changes. Methodology 0.3.7 does **not** trigger a bulk migration or auto-replace `?` with `—`. Reassessments (including suspected Index batch #1314 drift) require separate source evidence, existing Index review/admission/checkpoint gates, and preserved immutable `generated_*` provenance.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall