opensiro/vsm-harness-skills / skills/assess-vsm-harness
assess-vsm-harness
VerifiedProduce a standalone evidence-backed VSM assessment of one autonomous AI agent harness at a pinned repository revision, including first-party ownership modes.
| File | Kind | Size |
|---|---|---|
| CHANGELOG.md | markdown | 13.4 KB |
| SKILL.md | skill md | 17.8 KB |
| VERSION | other | 6 B |
| references/assessment-format.md | markdown | 14.6 KB |
| references/autonomy-states.md | markdown | 14.4 KB |
| references/interventions.md | markdown | 4.2 KB |
| references/negative-state-calibration.md | markdown | 5.7 KB |
| references/positive-evidence-calibration.md | markdown | 15.5 KB |
| references/positive-evidence-sample-2026-10-13.md | markdown | 7.4 KB |
| references/profile/PROFILE.md | markdown | 24.9 KB |
| references/token-compute.md | markdown | 2.2 KB |
references/assessment-format.md
View on GitHub# Standalone assessment format
**Methodology version:** 0.3.8
An assessment is repository-relative and revision-relative. It must not contain cohort-relative signatures, rank positions, or claims that depend on which other harnesses happen to be indexed.
Use flat frontmatter so deterministic index tooling can parse it without a YAML dependency:
```yaml
---
harness_id: example-harness
project_name: Example Harness
repository: https://github.com/example/harness
review_ref: <40-character commit SHA>
reviewed_at: YYYY-MM-DD
generated_profile_version: 0.2.4
generated_assessment_procedure_version: 0.3.8
profile_version: 0.2.4
assessment_procedure_version: 0.3.8
status: included
autonomy_s1: A
autonomy_s2: C
autonomy_s3: A(P)
autonomy_s3_star: ?
autonomy_s4: C(P)
autonomy_s5: P
---
```
`generated_profile_version` and `generated_assessment_procedure_version` record the versions that produced the original artifact. They are immutable origin metadata: once written, reassessment, correction, or Methodology migration must not rewrite them.
`profile_version` records the normative VSM Harness Profile under which the current classification was most recently successfully produced or revalidated. `assessment_procedure_version` records the VSM Harness Methodology version used for that current accepted classification.
The `*_assessment_procedure_version` names are retained as compatibility storage keys. From Methodology v0.2.2 onward they do not imply a separately versioned assessment-only procedure: the same Methodology release also governs synthesis, ranking projection, and validation.
For legacy assessments whose generation versions were never recorded, omit both `generated_*` fields rather than guessing. Generation provenance is optional only for such legacy artifacts; every assessment newly created by Methodology v0.2.1 or later must emit the pair.
Allowed final `status` values for canonical index entries are `included` and `excluded-no-agentic-vsm`. A downstream intake workflow may temporarily use its own pre-admission status outside the canonical assessment contract; do not confuse that lifecycle status with an accepted assessment state. For excluded rows, omit positive autonomy claims and explain the exclusion in the body.
## State notation
The allowed publication symbols are:
```text
A A(P) C C(P) P — ?
```
`A(P)`, `C(P)`, and standalone `P` are valid only for S3, S4, and S5 in Methodology `0.3.x`.
- `A(P)` means the autonomous `A` mode is established and a distinct first-party parent-governed mode for the same function is also operationally closed.
- `C(P)` means the constructor `C` mode is established and a distinct first-party parent-governed mode for the same function is also operationally closed.
- `P` means the parent-governed mode is the only positive ownership mode established at the reviewed boundary; no first-party `A` or `C` autonomous mode is established there.
The composite notation records supported first-party ownership modes, not simultaneous dual ownership. For each concrete mode used as evidence, the decisive right must still have a reconstructable owner and closure path.
The parent-mode boundary is intentional:
- S1 and S2 do not publish `P`; the Methodology is assessing autonomous agent harnesses rather than every possible human-owned organizational arrangement;
- S3 and S4 admit parent modes as explicit supervisory/current-control and adaptation exceptions;
- S3* does not publish the parent modifier in `0.3.x`;
- S5 is the canonical parent-governed case for identity / ultimate-policy authority.
This is a publication choice, not a claim that human-owned S1/S2/S3* cannot exist outside the assessed autonomous-harness boundary.
## Methodology 0.3.7 negative-state calibration note
Before `—` or `?`, check plausible function-specific first-party paths in the supported mode, including positive and parent-governed closure options. The existing three-part `Absence scope` may support a **bounded** conclusion that no material path is shipped (`—`); it is not a claim of universal impossibility. Use `?` when an unresolved plausible *shipped* path or inadequate inspection coverage prevents a defensible positive or negative assessment. Identify the actual gap under the existing `Evidence gaps` section when possible. No additional mandatory artifact field is introduced. See [negative-state-calibration.md](negative-state-calibration.md).
## Methodology 0.3.6 Profile-sync note
Methodology `0.3.6` synchronizes the bundled normative dependency to compatible Profile `0.2.4`. It introduces **no new assessment fields, ownership states, evidence thresholds, synthesis rules, or ranking semantics**. The boundary-provenance requirements introduced in `0.3.5` remain the current structural contract for newly produced or revalidated `0.3.6` assessments.
Existing assessments do not advance their current Methodology/Profile provenance merely because `0.3.6` exists; they advance only after normal successful revalidation. Immutable `generated_*` provenance is never rewritten.
## Methodology 0.3.5 boundary-provenance requirements
For every newly produced or successfully revalidated `0.3.5` or later compatible artifact, the `Review boundary` section must include:
```markdown
- Credited operating / distribution surfaces:
- Adjacent first-party surfaces excluded from ownership:
```
The first field names the shipped/runtime/installed surfaces that may supply credited owners and closure. The second records nearby first-party development/dogfood, contributor, CI/release, benchmark/evaluation, test/example or governance surfaces that were inspected but are outside the ownership boundary. Use `none identified` when the review found no material adjacent surface; do not omit the field.
Every positive function state (`A`, `A(P)`, `C`, `C(P)`, or `P`) must additionally include:
```markdown
- Boundary reachability:
```
This explains why the decisive actor or function-specific constructor path is reachable in the declared assessed mode without borrowing ownership or closure from an adjacent system. Same-repository location alone is not a reachability argument.
## Methodology 0.3.3 structural completion requirements
The following requirements make the existing evidence standard mechanically checkable without turning semantic judgment into a unit test.
For every `State: —`, include:
```markdown
### Absence scope
- Surfaces inspected:
- Plausible first-party paths checked:
- Why no material first-party path remains:
```
The three lines must describe a boundary broad enough to support a no-material-path conclusion. If that conclusion cannot be defended, use `?`.
For every `A(P)` or `C(P)`, include this table in the corresponding function section:
```markdown
| Mode | Decisive owner | Trigger | Closure | Evidence |
| --- | --- | --- | --- | --- |
| Base (`A` or `C`) | ... | ... | ... | ... |
| Parent (`P`) | ... | ... | ... | ... |
```
Use the actual base state in the first row (`A` for `A(P)`, `C` for `C(P)`). The table is evidence organization, not a new ownership state.
A `0.3.3` or later compatible artifact can be structurally checked with:
```bash
python scripts/check_assessment_contract.py assessments/<harness_id>.md
```
The checker is a completion oracle, not a semantic oracle. It cannot establish VSM function, ownership, independence, or evidentiary truth.
## Required body
```markdown
# <Project name>
## Review boundary
- System in focus:
- Purpose and identity:
- Relevant environment:
- Standard-distribution boundary:
- Credited operating / distribution surfaces:
- Adjacent first-party surfaces excluded from ownership:
- First-party operating / deployment modes considered:
- Recursion level:
- Reviewed revision:
- Observation date:
- Generated Profile version:
- Generated Methodology version:
- Current Profile version:
- Current Methodology version:
## Repository architecture
Describe enough of the repository to preserve the assessment context: runtime, agent loop, state, tools, delegation, persistence, scheduling, evaluation, human involvement, maintainer/contributor governance, supported self-hosted/operator modes, and external integration when material.
## Operational model
Identify the actual operational outcomes and S1 units. Separate agent decision rights from runtime, developer, human, parent, and distributed governance authority. When a composite state is claimed, describe the autonomous/constructor mode and parent-governed mode separately.
## S1 — Operations
- State:
- Function:
- Disturbance / variety regulated:
- Decisive decision or feedback right:
- Decision owner:
- Supporting / enforcement mechanisms:
- Closure path:
- Boundary reachability:
- Why this is / is not agent-owned:
- Evidence:
- Basis:
- Confidence:
- Caveats:
For `State: —`, additionally include:
### Absence scope
- Surfaces inspected:
- Plausible first-party paths checked:
- Why no material first-party path remains:
## S2 — Coordination
Use the same fields. For a positive S2 state, additionally include these exact witness labels:
- Distinct S1 units:
- Inter-S1 disturbance:
- Attenuating coordination relation:
- Feedback into subsequent S1 behaviour:
- Why this is S2-specific rather than generic communication / routing / sequencing / shared state / delegation:
For `S2=C`, the S2 function itself must already be established. `C` means a first-party S2-specific decision/feedback path exists but the autonomous actor, authority, or closure still requires composition; it does not mean a generic framework could be programmed into S2. Methodology `0.3.x` does not apply `(P)` or standalone `P` to S2.
## S3 — Inside-and-now control
Use the same fields. For a positive S3 state, additionally include:
- Whole-system current view:
- Current-control decision scope:
Explicitly distinguish resource/commitment decisions from deterministic budget, scheduler, concurrency, or termination enforcement.
For `S3=P`, `S3=A(P)`, or `S3=C(P)`, additionally establish the parent mode:
- whole-system current view at the declared recursion;
- current-control decision over resources, commitments, priorities, constraints, accountability, synergy, or intervention;
- legitimate parent human/institution/higher recursion/distributed parent arrangement as decision owner in that mode;
- returned decision changing subsequent current operation.
For `S3=A(P)`, separately establish the autonomous `A` closure. For `S3=C(P)`, separately establish the S3-specific constructor path that justifies `C`.
A local human intervention, generic operator UI, or merge approval alone is not sufficient for the `(P)` modifier.
## S3* — Complementary audit
Use the same fields. For a positive S3* state, additionally include:
- Claim being audited:
- Ordinary reporting path:
- Complementary access path:
- Independence boundary:
- Who acts on findings:
Methodology `0.3.x` does not apply the parent-mode modifier to S3*.
## S4 — Outside-and-then intelligence
Use the same fields. For a positive S4 state, additionally include:
- External distinction:
- Future / prospective distinction:
- Adaptation option generated:
- Path back into current capability / S3:
For `S4=P`, `S4=A(P)`, or `S4=C(P)`, additionally establish the parent mode:
- parent ownership of the decisive adaptation judgment/feedback right;
- return of that parent decision into present capability;
- subsequent operational/capability change under that returned decision.
For `S4=A(P)`, separately establish the autonomous `A` adaptation closure. For `S4=C(P)`, separately establish the S4-specific constructor path that justifies `C`.
A roadmap, research issue, backlog, or human idea alone is not sufficient for the `(P)` modifier.
## S5 — Policy and identity
Use the same fields. For a positive S5 state, additionally include:
- Identity / ultimate-policy issue:
- Ultimate authority in each claimed mode:
- Return-to-operation path:
For `S5=P`, `S5=A(P)`, or `S5=C(P)`, show the complete parent mode: identity/policy issue → legitimate parent authority → authoritative decision → decision returned → subsequent operation governed by it.
For `S5=A(P)`, separately establish a first-party mode in which the agent/system internally closes S5 and owns the decisive identity/ultimate-policy right. The notation describes alternative supported ownership configurations; it does not claim two simultaneous ultimate authorities.
For `S5=C(P)`, separately establish the S5-specific constructor path that justifies `C`.
## Distributed OSS parent arrangement
When the reviewed system is open-source and multiple independent contributors operate local agents without a shared private runtime context, record whether parent governance is local to one contribution cell or established at the organization/project recursion. Do not infer organization-level `(P)` merely from multiple human contributors. Positive parent-mode notation requires function-specific legitimate ownership and a reconstructable return/closure path at the declared boundary.
## Self-hosted and non-human modes
For self-hosted harnesses, inspect first-party operator/supervised modes when they materially alter S3/S4/S5 ownership. A supported operator mode can justify `(P)` only when the function-specific parent loop is complete; generic extensibility is not enough.
For intentionally non-human organizations, absence of `(P)` is not a deficiency. `S3=A`, `S4=A`, or `S5=A` may be correct when those functions close autonomously and no qualifying parent-governed mode is established.
## Recursion
## Variety and escalation
## Evidence gaps
```
`Closure path` may be `not applicable` when the function or negative finding does not require a distinct returned decision, but positive S2, S3, S3*, S4, and S5 claims should make the relevant feedback path explicit enough for a second reviewer to reconstruct the regulation loop.
## Evidence requirements
Prefer immutable primary-source permalinks at the reviewed revision. Evidence may include maintainer documentation, source, examples, architecture material, tests, governance artifacts, issues/PRs, releases, or observable traces. Positive `A`, `A(P)`, `C`, `C(P)`, or `P` states require enough cited evidence for a second reviewer to reproduce the function mapping and every ownership/closure mode encoded by the symbol.
A deterministic support mechanism is evidence that a decision can be operationally enforced; it is not, by itself, evidence that the mechanism owns the organizational decision. When the owner is ambiguous, record that ambiguity and apply the counterfactual owner test rather than inferring agent or parent ownership from the existence of hard enforcement.
`?` means evidence is insufficient. `—` requires a reviewed boundary broad enough to support the narrower claim that no material first-party path is supplied there; it is not a universal impossibility claim.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| 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: 472k | Updated | Claude CodeCodex | Verified | Install |
| 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: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| 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 available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |