Skip to content

Ascension Cycle

Jump to bottom
mbtiongson1 edited this page Jun 22, 2026 · 2 revisions

Ascension Cycle

How a skill travels from a developer's repo to the canonical graph.

Overview

Your Repo                    Registry
─────────────────────────────────────────────────────────
gaia scan          →  detects skills in your code
gaia push          →  registry-for-review/skill-batches/<id>.json  (pending)
                             │
                    Maintainer reviews issue
                             │
                       ┌─────▼──────┐
                       │  awakened  │   ← accepted into the graph at 1★
                       └─────┬──────┘
                             │
                    gaia propose <skillId> --target contributor/skill
                             │
                       ┌─────▼──────┐
                       │   named    │   ← contributor claims it (2★+)
                       └────────────┘
                             │
                    Evidence accumulates (G7 Trust Magnitude)
                             │
                    3★ → 4★ → 5★ → 6★ ★ (Apex)

Stage 1: Pending

When you run gaia push, Gaia writes a batch intake record to registry-for-review/skill-batches/ and opens a draft PR. Each proposed skill in the batch has "lifecycle": "pending".

The batch includes:

  • Detected canonical skills — skills already in the registry nodes that your code demonstrates.
  • Proposed new skills — capabilities detected that don't yet exist in the graph.

Stage 2: Awakened

A maintainer reviews the intake PR. Each proposed skill is classified as one of:

Decision Meaning
accept The skill is valid; promote it to registry/gaia.json at 1★
rename Accept with a different ID or name
duplicate The skill already exists under another ID
needs-evidence Concept is valid but the evidence bar isn't met yet
reject The skill is out of scope, vendor-specific, or not falsifiable

Accepted skills are added to registry/nodes/ with level: "1★" and status: "provisional". Their lifecycle field is updated to "awakened".

Stage 3: Named

Once a skill is awakened (1★), the contributor who discovered it (or anyone else) can publish a named implementation:

gaia propose web-search --target yourname/my-skill

Or interactively:

gaia propose

This creates registry/named/yourname/my-skill.md with YAML frontmatter. The first person to publish a named implementation for a skill bucket gets origin: true.

Named skills must be 2★ or above. 0★/1★ is reserved for generic, unclaimed skills.

→ See Named Skills for full details.

Stage 4: Rank Progression

After a skill is in the graph, its rank advances through evidence:

  1. Submit a reclassification PR (or calibrate level with gaia dev calibrate) with new evidence entries in the corresponding file under registry/nodes/ (or registry/named/).
  2. Add evidence via gaia dev evidence or direct JSON/frontmatter edits.
  3. CI validates the evidence type, grade, and level (rank) combination.
  4. A maintainer approves; the skill's level field is updated.

Under the G7 Trust Taxonomy model, evidence is evaluated on two axes:

  • Evidence Type (provenance of the demonstration)
  • Evidence Grade (S/A/B/C quality grade, derived from its trustNumber).

Ultimate skills (◆) require ≥ 5 prerequisites, Origin Fusion, and ≥ 10k repository stars. Reaching 6★ (Apex) requires passing the 6 active predicates of the G7 Apex gate.

Lifecycle Field Values

Value Where it appears Meaning
pending Intake batch JSON Proposed, not yet reviewed
awakened Intake batch JSON + registry/nodes/ Accepted by maintainer, 1★
named Intake batch JSON + registry/named/ Contributor has claimed the implementation

Diagram: Full Path for autonomous-debug 

Developer writes a debugger agent
       ↓
gaia push  →  batch-xyz.json
  { id: "autonomous-debug", lifecycle: "pending",
    prereqs: ["code-generation","execute-bash","error-interpretation"] }
       ↓
Maintainer: accept
       ↓
registry/nodes/extra/autonomous-debug.json updated:
  { id: "autonomous-debug", type: "extra", level: "1★",
    prerequisites: ["code-generation","execute-bash","error-interpretation"] }
       ↓
gaia propose autonomous-debug --target devin-ai/autonomous-swe
       ↓
registry/named/devin-ai/autonomous-swe.md
  { origin: true, level: "3★", genericSkillRef: "autonomous-debug" }
       ↓
Evidence PR / gaia dev calibrate / gaia dev evidence → 3★ → 4★ → ...

Clone this wiki locally