G∞
GoalOS REAL-001Live repository reliability transfer · fresh GitHub work · independent replay ready
The decisive demonstration · consequential real work

Mission 1 must make Mission 2 better.

This proof room tests verified capability on a live GitHub maintenance decision—not a synthetic score. A merged repository-stabilization mission produces a Chronicle-admitted dependency reliability skill. A later Dependabot pull request then asks for a major plugin upgrade that conflicts with the repository’s Hardhat baseline.

The test: same fresh task, same evidence universe, same decision classes, same observation budget, one patch proposal. Only the prior changes: none, raw memory, or a structured Validated Skill.

The real task

LOCAL EVIDENCE
Mission 1
ChronicleAdmit a scoped repository reliability skill.
Fresh Mission 2
Equal armsFresh · raw memory · Validated Skill · ungated ablation.
Independent replayTwo external operators must reproduce the roots.
AuthorityNo automatic merge. Human review remains mandatory.

The proof room.

Every assertion is inspectable. Every stronger claim remains gated.

Authority sequence

A stronger state appears only after the next gate passes.

READY

Measured transfer

The skill must improve the next mission—not merely be reused.

READY TO RUN

Press Run proof to reconstruct the campaign.

Equal constraints. Different memory.

The comparison controls for task, evidence universe, step budget, decisions, and patch allowance.

REAL PR SNAPSHOT

Mission Contract

What the work is allowed to decide.

Public Evidence Docket

What passed, failed, and remains unproven.

Validated Skill Passport

A capability object—not a transcript.

ADMIT_WITH_SCOPE

Why it helps

Mission 1 compresses proven repository-specific search into an indexed prior.

BaselineExact core dependency versions
RulesMajor / peer mismatch migration gates
TestsKnown safe validation route
Memory law: raw notes may hint. The skill carries scope, evidence, compatibility rules, validation commands, freshness, and rollback.
Useful because it is bounded.

The skill cannot merge, deploy, move tokens, use secrets, or claim production authority.

Committed state

A single mutation invalidates the skill’s authority.

VERIFYING

Integrity checks

Commitment is not truth; it is exact-state integrity.

VERIFYING

Launch on a live repository

Generate the exact workflow command for a fresh GitHub task.

Autonomous safety architecture

Useful automation without granting hidden authority.

NO AUTO-MERGE
Trusted selection
Selects a live protected Dependabot PR from the default branch.
PASS
Read-only proof job
No pull-request code receives write credentials.
PASS
Artifact attestation
GitHub OIDC/Sigstore provenance for Proof Packs and receipts.
PASS
Human-review PR
The write job opens only an Evidence Docket PR from trusted code.
PASS
Independent replay
Two distinct operators remain open Proof Debt.
PENDING