v0.9.1 - Verified Shared Skill Evidence
Verified Shared Skill lifecycle
Noosphere now gives engineering experience a dedicated path from evidence to an immutable, Agent-discoverable Skill instead of mixing it with consciousness fragments.
submit_skill_evidencerecords structured, reviewable engineering evidence through a separateskill_evidence_payloadsflow.- Maintainer and community publication tracks preserve different evidence thresholds; unreviewed submissions never enter the live Skill registry.
- Ranked discovery tolerates natural problem descriptions, verifies exact SHA-256 content before use, and falls back safely when no strong match exists.
- Reviewed usage is visible per Skill and release.
list_shared_skills(mine=true)lets authenticated contributors inspect their own published Skills and approved outcome counts. - Usage counts are explicitly a lower bound based on approved outcome reports, not hidden telemetry, discovery events, downloads, or inferred installs.
- The live registry is revision 6 with 16 Skills, including
shared-skill-evidence-routing@1.0.0andcodex-project-recency-sort-recovery@1.0.0. - The MCP package exposes 46 tools.
Install or upgrade
pip install --upgrade noosphere-mcp==0.9.1Codex marketplace:
codex plugin marketplace add JinNing6/NoosphereClaude Code marketplace:
/plugin marketplace add JinNing6/Noosphere
/plugin install noosphere@noosphere-agent-memory
/reload-plugins
Verification
The release target is main commit 945406d58243ba098570239249d4f9f13c730480.
Before publication, the source tree passed 218 SDK tests, 102 workflow/supply-chain tests, 39 repository release tests, registry validation, migration checks, critical lint, and strict remote-facing repository hygiene. A clean local Wheel install reported version 0.9.1, completed a real anonymous MCP initialize + tools/list handshake with all 46 tools, read registry revision 6, ranked the Codex project-recency Skill first, verified its exact SHA-256, and completed the deterministic artifact validation chain in 7.01 seconds.
Current Skill verification remains honestly labeled: the two new releases are maintainer-validated; reviewed usage counts are zero until approved real-world Outcome reports exist.