Repository navigation
Released 2026-09-15.
Humanities Superpowers v2.1.0 promotes the verified Dual-Core work into a release. It is not a new research feature: the Humanities/Fricturn research core keeps its 13 core skills, its router semantics, its verification states, its gate statuses, its validity rules, and its human authorization boundary.
What v2.1.0 contains
One router, two cores, and an explicit boundary between them.
Router
|
-----------------------------
| | |
RESEARCH CODE HYBRID
| | |
Humanities Engineering Both cores
Core Core
Added
- Dual-Core architecture — a research core and an engineering core reached through one router, with an explicit domain boundary.
- Engineering core — ten selected
obra/superpowers6.3.0 skills, vendored byte-identically with per-skill hashes, import commit, license, and a documented reason for every excluded skill. RESEARCH/CODE/HYBRIDrouting — semantic classification by the artifact that must be correct, not by keywords.QUICK/STANDARD/STRICTrisk routing — applied to code and hybrid work, derived from reversibility, blast radius, privilege boundary, persistence, contract surface, environment, failure cost, and uncertainty.- Autonomous Worker state contract —
WORKING,WAITING_INPUT,WAITING_PRIVILEGE,BLOCKED,ERROR,DONE, where a live process is notWORKINGand a terminated process is notDONE. WORK_PACKAGE/EVIDENCE_PACKAGE— handoff records between role-basedPLANNER,IMPLEMENTER,REVIEWER, andESCALATION_REVIEWERassignments, with model names confined to a replaceable operating profile.- Coding evaluation harness —
scripts/hsp_eval_coding.pyandtests/eval/coding-cases.jsonfor the separateCONTROL_CODINGversusENGINEERING_COREcomparison.
Preserved
- The 13 core research skills are unchanged from the audited tip (
2a452b6) apart from the single release-metadataversion:line; the skill bodies, gates, and validity rules are identical. - Gate status, current validity, and progression authorization stay separate properties, and invalidation remains bounded by dependency references.
- Epistemic return, citation verification, productive refusal, terminology control, and human judgment authority are unchanged.
- Installation safety: an existing
AGENTS.mdorCLAUDE.mdis never overwritten, andskills/skills/is never created. vendor/obra-superpowers/**is byte-identical to upstream commitb36e0829c6d0140e93cfef2ca599b1b07d4a7797.
Limitations
- Coding effectiveness:
NOT_MEASURED. NoCONTROL_CODINGversusENGINEERING_CORErun has been recorded, so this release claims no coding-quality improvement, no defect reduction, and no token saving. - Research effectiveness:
NOT_MEASURED. NoCONTROLversusHSPrun has been recorded, so this release claims no research-quality improvement. - Runtime notification: not implemented. The worker state contract is a contract. No Telegram, Hermes, or other notification runtime ships with this release.
- Cursor runtime loading:
NOT_MEASURED. The Cursor project layout is checked statically, but the rule has not been loaded in a target Cursor environment. - Passing repository, installation, and conformance checks demonstrate internal consistency. They are not evidence of benefit.
Version scope
v2.1.0 is the project release version. It is not the Fricturn protocol version: the v2 specification documents remain at Specification Version: 2.0.0, schema identifiers are unchanged, and schema_version values in protocol records are unchanged.
Provenance
| Field | Value |
|---|---|
| Upstream repository | https://github.com/obra/superpowers |
| Upstream version | 6.3.0 |
| Upstream commit | b36e0829c6d0140e93cfef2ca599b1b07d4a7797 |
| Upstream license | MIT, Copyright (c) 2025 Jesse Vincent |
| Imported skills | 10 of 14 |
| Record | vendor/obra-superpowers/PROVENANCE.json |
The vendored files are unmodified. scripts/run_dual_core_tests.py recomputes the recorded hashes and fails if any vendored byte changes.
Validation
The release candidate passed repository validation, integrated tests, v2 conformance tests, dual-core routing and provenance tests, public-release validation, installation smoke tests, both evaluation self-tests, and a strict documentation build. Manifest hashes were regenerated from the release tree.
Codex and Claude Code clean installs were verified: the documented commands were executed in isolated projects and reproduced 14 Humanities skills, 10 vendored engineering skills, the router, templates/, schemas/, the dual-core specification, and the provenance record, without overwriting an existing AGENTS.md or CLAUDE.md and without creating skills/skills/.
The vendor tree, manifest, and the v2.1.0 tag were re-verified on main after the fast-forward merge, and the same validation suite passed again there.