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.

CHANGELOG.md

View on GitHub
# assess-vsm-harness changelog

## 0.3.8 — 2026-10-13

- publish informative, source-pinned R1–R10 positive-evidence calibration cases from independent external harnesses, with an explicit 20-recent-admissions / 12-targeted / 23-distinct sample manifest;
- illustrate existing tests for supported runtime entry/binding/guard, function-specific ownership and feedback, S3* complementary access, S3 current discretion, S4 capability adaptation and concrete evidence/event identity;
- preserve Profile `0.2.4`, unchanged `A/A(P)/C/C(P)/P/—/?` thresholds, assessment fields, provenance, synthesis/ranking and Index canonical grades; no automated migration or blind/independent adjudication claim;
- extend the existing structural completion checker to recognize `0.3.8` without introducing new completeness requirements.

This is a backward-compatible, calibration-only Methodology patch. The sample is nonrandom, contested examples remain unresolved until Index-owner disposition, and original `generated_*` fields and frozen study inputs are not rewritten.


## 0.3.7 — 2026-10-10

- clarify bounded negative-state adjudication without weakening the existing three-part absence-scope requirement;
- add an explicitly non-normative calibration guide with six immutable canonical Index comparison anchors;
- distinguish *unresolved plausible shipped paths* from hypothetical extensions, and require reviewers to check positive/parent modes first;
- extend Methodology structural checker version compatibility and add contract-regression tests while retaining exactly the same schema/evidence fields and thresholds;
- preserve Profile `0.2.4`, autonomy symbols, parent modes, synthesis and ranking semantics.

This is a patch-level clarification, not an automatic migration or a presumption that any published `?` is incorrect. Separate evidence-backed same-ref Index corrections retain immutable generation provenance.


## 0.3.6 — 2026-09-22

- sync the bundled normative dependency to compatible VSM Harness Profile `0.2.4`;
- carry the Profile clarification that algedonic signalling is an exceptional signal/channel path rather than an additional S-function or a scheduler/status state;
- preserve the assessment schema, ownership notation, evidence thresholds, synthesis procedure, and ranking projection unchanged;
- extend the structural completion oracle so `0.3.6` artifacts retain the same boundary-reachability contract introduced in `0.3.5`.

This is a patch-only normative-dependency/provenance update. Profile `0.2.4` declares `compatibility: compatible` and `assessment_impact: none`. No corpus-wide reassessment or vector migration follows solely from this release. Existing assessments advance current Profile/Methodology provenance only after normal successful revalidation, and immutable generation provenance is never rewritten.

## 0.3.5 — 2026-09-19

- sync the bundled normative dependency to compatible VSM Harness Profile `0.2.3`;
- make boundary provenance reconstructable by separating credited operating/distribution surfaces from adjacent first-party development/dogfood/test/eval/governance surfaces;
- require every positive function state to record `Boundary reachability`;
- clarify that same-repository dogfood can corroborate a constructor path but cannot supply ownership/closure to another system-in-focus merely by co-location;
- extend the structural completion oracle and regression contract for `0.3.5` while preserving historical `0.3.3`/`0.3.4` validation.

This is a patch-level procedural clarification. It does not redefine VSM functions, ownership symbols, parent-mode semantics or ranking projection, and it does not itself require corpus-wide reassessment. Weak historical mappings exposed by the clarified boundary are same-ref corrections under their frozen contracts.

## 0.3.4 — 2026-09-18

- fix `scripts/check_assessment_contract.py` so `## S3*` is parsed as the complementary-audit section rather than as a second `S3` section;
- replace the word-boundary section matcher with an explicit heading-boundary lookahead that works for the non-word `*` suffix;
- add a regression test covering all six function headings and distinct `S3` / `S3*` keys;
- allow the corrected oracle to validate `0.3.3` artifacts because `0.3.4` changes validator implementation only, not the `0.3.3` structural assessment contract;
- preserve Profile `0.2.2`, autonomy notation, evidence requirements and ranking semantics unchanged.

## 0.3.3 — 2026-09-18

- sync the bundled normative dependency to compatible VSM Harness Profile `0.2.2`, whose release declares `assessment_impact: none` and does not change S1–S5 semantics or evidence thresholds;
- require every newly produced or revalidated `—` to record an explicit absence scope: surfaces inspected, plausible first-party paths checked, and why no material first-party path remains at the declared boundary; use `?` when that negative conclusion cannot be defended;
- require `A(P)` and `C(P)` assessments to include a compact base-mode / parent-mode reconstruction matrix with decisive owner, trigger, closure and evidence for each mode;
- add `scripts/check_assessment_contract.py` as a structural completion oracle for `0.3.3` assessment artifacts; it checks required evidence surfaces but does not attempt to decide semantic correctness;
- replace temporary one-shot release publishers with a persistent changelog-driven publisher triggered by Methodology version changes;
- backfill missing public release tracking for Methodology `0.2.0`, `0.3.0`, and `0.3.1`, and add scheduled release-tracking validation so future versions cannot silently remain untagged/unreleased;
- preserve the existing ownership symbols, function-first classification, parent-mode semantics, ranking projection, and generation-provenance rules.

This is a patch-level Methodology reproducibility/release-hygiene change. It does not by itself require vector changes for already conforming assessments. Existing assessments advance their current Methodology provenance only after normal successful revalidation. Version `0.3.2` was not published; `0.3.3` is the next maintainer-selected patch version.

## 0.3.1 — 2026-09-16

- clarify the intended publication boundary of parent-governed notation without changing the state set introduced in `0.3.0`;
- keep `P`, `A(P)`, and `C(P)` intentionally limited to S3, S4, and S5 rather than treating parent ownership as a universal modifier for every VSM function;
- define S5 as the canonical parent-governed case because identity / ultimate-policy authority naturally resides at a legitimate parent recursion when it is not internally agent-owned;
- define S3 and S4 parent modes as explicit supervisory / adaptation exceptions that are useful for self-hosted, operator-assisted, and distributed OSS systems where current-control or adaptation rights may intentionally remain parent-owned;
- state that S1 and S2 do not receive `P` in this Methodology: the assessment target is an autonomous AI agent harness, so ordinary operational ownership and inter-S1 coordination are expected to be agent-owned or constructor paths rather than published as human-owned harness modes;
- keep S3* outside the parent-mode notation in the `0.3.x` line; audit classification remains focused on complementary access, independence, audit judgment, and feedback rather than adding a parent modifier;
- distinguish descriptive possibility from Methodology scope: human ownership can exist in real organizations outside these publication choices, but the Methodology intentionally records only ownership topologies useful for assessing autonomous agent harnesses;
- preserve all `0.3.0` ranking, parsing, migration, and provenance semantics.

This is a patch-level Methodology clarification. It does not change VSM Harness Profile semantics, does not add or remove any publication symbol, and does not by itself require a new reassessment outcome for an already correctly reviewed `0.3.0` assessment. R2 should apply the clarified boundary while reviewing S3/S4/S5 ownership modes.

## 0.3.0 — 2026-09-16

- keep VSM Harness Profile `0.2.1` as the normative semantic dependency; this release does not redefine S1–S5, S3*, recursion, autonomy, or variety;
- extend the Methodology publication notation for S3, S4, and S5 from mutually exclusive `A/C/P/—/?` states to support composite first-party ownership modes `A(P)` and `C(P)`;
- define `A(P)` as an established autonomous `A` mode plus a distinct operationally closed parent-governed mode for the same function;
- define `C(P)` as an established constructor `C` mode plus a distinct operationally closed parent-governed mode for the same function;
- retain standalone `P` for S3/S4/S5 when a parent-governed mode is operationally closed but no first-party `A` or `C` autonomous mode is established at the reviewed boundary;
- require each encoded ownership mode to be independently reconstructable from primary evidence; composite notation never means simultaneous dual ownership of one decisive right in one concrete deployment/run;
- make self-hosted/operator modes and distributed OSS parent arrangements explicit assessment evidence surfaces without inferring organization-level parent governance from generic human involvement;
- explicitly allow non-human organizations to remain plain `A` for S3/S4/S5 when those functions close autonomously and no qualifying parent-governed mode is established;
- preserve the ranking key on agent-owned coverage by treating `A(P)` as base `A` and `C(P)` as base `C`; parent-mode presence remains descriptive and unweighted;
- require downstream Index tooling to understand the composite notation before assessments using it are admitted.

This is a substantive Methodology classification change. Existing canonical assessments do not automatically acquire `(P)` from historical mentions of humans, maintainers, approvals, or self-hosting. Reassessment must first re-establish the relevant S3/S4/S5 function and then independently establish the parent-governed closure at the declared recursion. Existing generation-origin metadata remains immutable; successfully migrated assessments advance only their current `assessment_procedure_version` to `0.3.0`.

## 0.2.3 — 2026-09-16

- sync the bundled normative dependency to VSM Harness Profile `0.2.1` at exact source revision `e1aaff7d2cd50d5d5ed9ab76c3606a6ef39d1976`;
- make the existing S2 evidence threshold mechanically reconstructable in the assessment workflow and artifact format;
- require positive S2 mappings to identify distinct S1 units, a specific actual or structurally evidenced inter-S1 disturbance, the coordination relation that attenuates it, and feedback into later S1 behaviour;
- clarify that `S2=C` requires an already-established S2 function plus a first-party S2-specific decision/feedback path; generic communication, routing, shared state, sequencing, or extension points remain insufficient;
- keep ownership separate from function evidence so deterministic coordination support does not become agent-owned `A` by enforcement alone.

This is a procedural clarification aligned with Profile `0.2.1`. It does not redefine the `A/C/P/—/?` states. Existing canonical assessments that violate the already-frozen Profile `0.2.0` / Methodology `0.2.1` S2 rule may be corrected as same-ref corrections rather than treated as migrations caused by this release.

## 0.2.2 — 2026-09-16

- define the existing `0.2.x` release line as the unified VSM Harness Methodology version;
- make assessment, cohort-relative synthesis, deterministic ranking projection, and validation part of one procedural release boundary;
- keep the existing `assessment_procedure_version` and `generated_assessment_procedure_version` field names as compatibility storage keys whose values identify the Methodology version when applied to assessment artifacts;
- explicitly avoid independent semantic versions for synthesis, ranking, TLDR, or RANKINGS while they are released as one methodology and materialized in a Git-versioned Index state;
- move the ranking projection rule into the Methodology contract without changing the current ranking output.

This is a release-boundary/provenance clarification. It does not change VSM Profile semantics or any `A/C/P/—/?` classification rule.

## 0.2.1 — 2026-09-16

- add immutable generation-origin provenance for newly created assessments;
- distinguish `generated_profile_version` / `generated_assessment_procedure_version` from the current accepted `profile_version` / `assessment_procedure_version`;
- require new v0.2.1+ artifacts to preserve their generation pair while allowing current semantic provenance to advance through reassessment;
- keep legacy origin unknown rather than fabricating historical generator versions.

This is a provenance/schema patch. It does not change VSM function semantics or autonomy-state classification rules.

## 0.2.0 — 2026-09-16

- align the procedure with VSM Harness Profile v0.2.0;
- separate organizational function, decisive decision/feedback right, owner, supporting enforcement, and closure;
- tighten `A` around agent ownership of the decisive organizational discretion;
- tighten `P` around complete parent-governed S5 return-to-operation closure;
- add the counterfactual owner test;
- add Profile/procedure provenance fields to the version-aware assessment format;
- resync the bundled Profile and record its immutable source revision.

The pre-0.2.0 procedure was not explicitly versioned. Do not fabricate a historical procedure version in existing assessments; record `0.2.0` when an assessment is first successfully reviewed under this procedure.
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: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated 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: 323kUpdated Claude CodeCodexVerifiedInstall