agent-forge
Design an agent only when a durable autonomy boundary is justified; compile it only when a Foundry runtime is available.
Skill by Arcanea-Labs
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/arcanea-labs-agent-forge-e827e6
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: design, backend, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions: not checked yet.
What's inside
- Skill name
agent-forge- Description
- Design an agent only when a durable autonomy boundary is justified; compile it only when a Foundry runtime is available. Use for persistent decision rights, distinct memory, constrained tools, ownership transfer, or an ongoing schedule, channel, or API trigger.
- License
- MIT
- Instructions
- 68 lines in SKILL.md · view on GitHub
Files (3)
- SKILL.md
- openai.yaml · agents/
- portable-contracts.md · references/
SKILL.md (first lines)
# Agent Forge
## Outcome
Design the smallest justified persistent actor, or deliberately downgrade the request to a skill or temporary worker. Compile it only when a Foundry runtime is present.
## Agent necessity gate
An agent must own at least one durable boundary:
- stable decision rights;
- a distinct memory scope;
- a constrained tool or permission boundary;
- genuine ownership transfer or handoff authority;
- an ongoing schedule, channel, or API trigger.
If none applies, stop and route to `$skill-forge`. A name, persona, or desire for “more intelligence” is not an agent boundary.
## Procedure
1. Build the Task Envelope.
- Record the objective, deliverables, stakes, autonomy, permissions, evidence policy, and termination tests.
2. Prove necessity.
- Mark each necessity-gate field true or false.
- Explain the persistent boundary in operational language.
- Compare the proposed agent against a skill, deterministic tool, and temporary task worker.
3. Define decision rights.
- State what the agent may decide, what it may recommend, and what requires operator approval.
- Exclude public release, financial commitment, destructive action, permission expansion, and substrate change unless explicitly authorized.
4. Define memory and tools.
- Name readable and writable scopes separately.
- Choose turn, session, or durable retention.
- Allow only the minimum tools needed; list denied tools explicitly.
5. Define handoffs and termination.
- Give every handoff a condition and target.
- Set success conditions, stop conditions, and a hard turn or round ceiling.
6. Compose existing skills.
- Agents orchestrate capabilities; they do not duplicate skill procedures inside a persona prompt.
7. Select the execution mode.
- **Foundry runtime available:** only when `tools/foundry/cli.mjs` and `foundry/contracts/` exist in the current workspace, compile and prove:
```bash
node tools/foundry/cli.mjs forge \
--envelope <task-envelope.json> \
--pack <agent-pack.json> \
--out <package-directory>
node tools/foundry/cli.mjs prove <package-directory>
```
- **Portable skills-only install:** read `references/portable-contracts.md`, then author the Task Envelope and Agent Pack with those field names. Mark compilation and proof `pending-runtime`; do not invent an Evidence Receipt. Provide the commands above as an optional handoff for a later SIS workspace run.
8. Refuse false persistence.
- If the boundary disappears during design, return a Skill Pack proposal instead of forcing agent creation.
## Quality gates
- At least one durable boundary is true and explained.
- Decision rights do not exceed the Task Envelope autonomy.
- Tool and memory permissions are least-privilege.
- Handoffs do not form an unbounded loop.
- The agent has explicit success and stop conditions.
- Behavioral, security, and drift evidence are required before production autonomy.
## Return
Return the necessity verdict, chosen topology, Agent Pack, execution mode, permissions requiring approval, and any downgrade recommendation. Include a compiled package path and Evidence Receipt only when they were actually produced.From Arcanea-Labs/Starlight-Intelligence-System, MIT.
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Sep 6, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
Same skill elsewhere
Copies of this SKILL.md published by others; search shows only one of them.
- 0.585score
agent-forge
frankxai
Design an agent only when a durable autonomy boundary is justified; compile it only when a Foundry runtime is available.
SkillAFreshness: A★ 11Claude Code +9 - 0.340score
agent-forge
Arcanea-Labs
Design an agent only when a durable autonomy boundary is justified; compile it only when a Foundry runtime is available.
SkillAFreshness: AClaude Code +9 - 0.329score
agent-forge
frankxai
Design an agent only when a durable autonomy boundary is justified; compile it only when a Foundry runtime is available.
SkillBFreshness: B★ 11Claude Code +9 - 0.329score
agent-forge
frankxai
Design an agent only when a durable autonomy boundary is justified; compile it only when a Foundry runtime is available.
SkillBFreshness: B★ 11Claude Code +9
Cite this page
agent-forge skill. Agent Mart. https://agentmart.dev/skills/arcanea-labs-agent-forge-e827e6 (last updated 2026-10-24)