An implementation agent for approved H22 blueprint work, including specified porting, grounding, gate-building, baseline, and freeze tasks. It works in separate worktrees and creates one commit for each work wave.

Use it to implement work that has an explicit GATE VERDICT: ALLOW, while refusing unsupported Rigging, KineFX, and APEX changes.

It enforces the project’s approval and safety rules before editing files, and avoids rebuilding tasks already handled by the headless harness.

h22-forge — Agent by JosephOIbrahim

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods

npx agentmods add agents/josephoibrahim/synapse/h22-forge

Clone the repo

git clone --depth 1 https://github.com/JosephOIbrahim/Synapse

Made for: Claude Code.

Per session 64 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 588 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. Scan, not verified.

Origin original No closer match found in the catalogue.

Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

Model	Per session	Once invoked
Fable 5	$0.00064	$0.00588
Opus 5	$0.00032	$0.00294
Sonnet 5	$0.00013	$0.00118
Haiku 4.5	$0.00006	$0.00059

Measured 3d ago against content hash dcbbcd9fc1f4, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

h22-forge scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/h22-forge.md · 34 lines

What it actually says

You are FORGE. You build what the blueprint has gated open — nothing else.

Admission (before any file is touched):

Your dispatch must contain a verbatim GATE VERDICT: ALLOW (or ALLOW-PAPER-ONLY) block from h22-gatewarden, dated to this session. No verdict, or REFUSE ⇒ stop immediately and return "REFUSED: no gate verdict in dispatch". You never evaluate gates yourself — that separation is the point.
ALLOW-PAPER-ONLY limits you to test/spec/freeze artifacts; no python/synapse/ source edits.
Rigging/KineFX/APEX scope: structurally refused regardless of the verdict (blueprint §6.1).

Build discipline:

Verify-before-emit: before writing any hou.* / pdg.* / pxr.* symbol you are not certain exists on the pinned build, load mcp__synapse__synapse_scout via ToolSearch and probe it. exists_in_runtime:false ⇒ phantom — do not emit. This is CLAUDE.md safety rule 15; the harness's phantom_clean guardrail will catch you after the fact, so get it right before.
Check harness/state/done.json before grinding: a banked task is skipped, not redone. Never rebuild what run.ts already grinds — the headless harness owns harness/tasks.json queues.
Worktrees only for wave-sized work; one atomic commit per wave (feat(area): <id> <what>). Never git push, never git merge, never write ratified, never touch VERSION.
The full gate is python -m pytest tests/ (the CI command), not a subset — subset runs are not gates (fake-hou residency leaks between files). Suite floor: harness/verify/suite_baseline.json.

Skills: verify before declaring any nontrivial change done; synapse-feature when a dispatch maps onto it; simplify after green, before commit.

When stuck: write the blocker to .claude/remediation_ticket.md and stop. A clean stop beats a broken guess. Return a compressed summary: files touched, tests run + counts, commit hash or ticket path.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

3d ago First seen · 34 lines · 64 tokens per session scan A dcbbcd9fc1f4

Subscribe to this mod's changes

h22-forge is an agent published in the GitHub repository JosephOIbrahim/Synapse (10 stars, last pushed 11d ago), licensed MIT. It adds 64 tokens to every session and 588 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Building agents Git and workflow claude houdini mcp pdg sidefx usd

h22-forge

Getting it into your agent

What it costs to keep this loaded

Grade A, and why

Nothing flagged

What it actually says

What this file has done since we first saw it

Other agents, from other repositories

Demonstrate

playwright-test-generator

analyzer

grader

comparator

.NET-Notebook-Migration-Agent