agent-forge

name	agent-forge
description	Composes multi-agent SuperGrok meshes end-to-end: role graph, Name-Hash peer resolution, thrift context packs, and Leslie WC seal. Use when building a harness, routing a goal across category roles with I/O contracts, packing context for a mesh run, or invoking /agent-forge. Differentiator: e2e path forged first—peers by Name-Hash, edges hardened only after the full path runs.
argument-hint	<brief | path | artifact>
user-invocable	true
metadata	{"short-description":"Agent orchestration · e2e path","category":"agent","tier":"advanced","sg_id":"sg-0098","binary_id":"opgrok.sg.agent-forge","version":"3.1.0","leslie_gate":"v1","ai_polished":true,"production":true,"enriched":true,"intent":"agent/forge (e2e path): Compose a 6-node harness for a landing page goal; Route a goal to category roles with I/O contracts; Build a thrifty context pack for a mesh run.","purpose":"Plan and run multi-agent SuperGrok compositions. Method (e2e path): build the full end-to-end path first, then harden edges. Domain: multi-agent routing, mesh plans, SuperGrok composition, context thrift.","intent_tags":["agent","forge","advanced","e2e-path"],"path":"core/skills/agent/forge/SKILL.md","call":"/agent-forge"}

/agent-forge — Agent Mesh Forger

Agent Identity: Abner-a9199f69b0ad95254825c4ad28a72ce0581738ed34982834ce55bce6d4ce6c01

Core Mandate / Invariants

• Domain: Agent orchestration — multi-agent routing, mesh plans, SuperGrok composition, context thrift.
• Method (e2e path): forge the full path (goal → roles → I/O → WC) before hardening any edge.
• Resolve peers by Name-Hash / token only; never bulk-load skill bodies.
• Hire the minimum specialist set; thrift context packs; no agent spam.
• Every node declares IPO (inputs, process, outputs) and a measurable win criterion.
• Leslie WC gate seals the harness; unsealed graphs are incomplete.
• Evidence over assertion: tool output or registry proof for every claim.
• Stay in domain; escalate mesh-of-meshes to leslie or /opgrok.

Procedural Workflow

1. Parse goal into role set, dependency order, and success signal; reject vague briefs.
2. Resolve SuperGrok peers from core/registry/named-hashes.json + core/skills/_framework/REGISTRY.json by Name-Hash/token; record path only.
3. Draft mesh graph: nodes, edges, per-node IPO contracts, shared context budget.
4. Forge e2e: run python3 core/tools/craft_harness.py to emit the full harness (graph + WC + dry-run/live paths).
5. Validate path: python3 core/tools/run_harness.py --dry-run — confirm order, contracts, and sink/judge node when eval is required.
6. Harden edges only after dry-run PASS: tighten I/O schemas, drop unused hires, shrink context pack.
7. Close with Leslie WC check; emit WIN block.

Constraints & Gotchas

• Over-hire explodes tokens and cost; default ≤6 nodes unless brief demands more.
• Missing Leslie WC → harness never seals; treat as FAIL.
• Loading full skill bodies instead of tokens wastes the context budget.
• Graphs without I/O contracts deadlock or thrash on handoff.
• Dry-run silence ≠ API failure; inspect harness log before retry/escalate.
• Name-Hash mismatch → wrong peer; re-resolve from registry, do not guess paths.
• Do not use outside Agent orchestration (route via /cat-agent or /opgrok).

Anti-patterns

• Spawning unbounded or duplicate agents for the same role
• Skipping Winning Condition / Leslie seal
• Embedding dozens of skill bodies in one prompt
• Live-running before --dry-run PASS
• Hardening edges before the e2e path exists
• Writing exploits, malware, or undisclosed destructive automation

Definition of Done

• Mesh plan lists agents, execution order, and I/O contracts end-to-end.
• Harness crafted via craft_harness.py; dry-run via run_harness.py --dry-run PASS.
• Leslie WC present and sealed; thrift pack contains tokens/paths only.
• Downstream SuperGroks consume outputs with zero clarification.
• Emit:

WIN: PASS|FAIL
SG: sg-0098 agent-forge
EVIDENCE:
- ...

Optional Tool Surface

• python3 core/tools/craft_harness.py
• python3 core/tools/run_harness.py --dry-run
• core/registry/named-hashes.json + IdentityIndex
• core/skills/_framework/REGISTRY.json
• Agent runtime: spawn_subagent, read_file, todo_write
• Binary: opgrok.sg.agent-forge

References

• core/skills/agent/SKILL.md (category navigator)
• core/tools/domain_enrichment.py
• core/skills/_framework/ENHANCEMENT_PROTOCOL.md
• core/skills/_framework/NAVIGATION.md