forge-agent

A command that turns a short description into a reusable agent system prompt. The result defines the agent's role, goal, working method, safeguards, output format, and when to challenge the requester.

In plain words
What is it for?
Use it to create prompts for subagents, custom GPTs, skills, or other systems that accept agent instructions.
Why use it?
It saves developers from designing those prompt parts from scratch and makes the agent's expected behavior explicit.

Command

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 commands/emtcmca/promptsmith/forge-agent
Clone the repo
git clone --depth 1 https://github.com/emtcmca/promptsmith
Per session 31 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,127 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.

ModelPer sessionOnce invoked
Fable 5 $0.00031 $0.01127
Opus 5 $0.00015 $0.00563
Sonnet 5 $0.00006 $0.00225
Haiku 4.5 $0.00003 $0.00113

Measured 2d ago against content hash c6828d604408, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

forge-agent 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 2d 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.

commands/forge-agent.md · 78 lines

How it starts

The opening of the file, as written. The whole thing — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Turn a short description into a complete, reusable agent system prompt. Output is plain text you can drop into a subagent, a skill, a custom GPT, or any system-prompt field.

Step 1 — Load the engine

Read ${CLAUDE_PLUGIN_ROOT}/skills/prompt-engineering/SKILL.md in full. This command runs the FORGE path.

Paths. ${CLAUDE_PLUGIN_ROOT} is this plugin's install directory, substituted automatically — never a literal folder in the user's project. If promptsmith was installed standalone (README Option B, no plugin root), read from ~/.claude/ instead: ~/.claude/skills/…, ~/.claude/promptsmith-templates/, ~/.claude/promptsmith-lenses/, ~/.claude/promptsmith-agents/. Never resolve these against the user's working directory.

Step 2 — Parse arguments

Parse $ARGUMENTS:

  • --lens <a,b> — lenses to bake into the agent's standing behavior. If absent, auto-pick the lens(es) that match the agent's domain (e.g. an editor agent → editorial).
  • --deep — interview one question at a time instead of assuming (engine Step 7).
  • Everything else = the description of the agent to build.

If the description is empty, ask what agent to build and stop.

Before building cold, check the gallery at ${CLAUDE_PLUGIN_ROOT}/agents/ for a specialist whose role is close to the request (the roster index is ${CLAUDE_PLUGIN_ROOT}/docs/agent-gallery.md). If one matches, read it and adapt it — swap domain, tone, and lenses to fit the description — rather than starting from a blank page. If nothing is close, forge fresh. Either way, the output meets the same bar.

Step 4 — Run the engine

Execute engine Steps 2–7 with route = FORGE. Key differences from SHARPEN:

  • Durable, not one-off. Write for every future run, not one task. Avoid task-specific detail.
  • Bake in the lens. The selected lens(es) become standing operating principles, not a one-time pass — the agent should always think like that professional.
  • Bake in push-back. Include an explicit self-challenge step in the agent's Method so it red-teams its own output before responding.
  • Define an output contract. The agent's responses must have a consistent, named shape.
  • Bake in the honesty floor (mandatory — every agent, no exceptions). Every forged agent's Constraints MUST include an explicit no-fabrication rule scoped to its domain: never invent facts, sources, quotes, citations, statistics, names, or claims; never assert a user-supplied fact as verified (attribute it as unverified, placeholder it, or decline — especially for legal, financial, regulatory, health, or safety claims); flag what's unconfirmed; and declare-and-degrade when a needed tool (retrieval, registry, data) is unavailable. This clause ships in 100% of forged agents — it does NOT depend on the red-team pass happening to surface fabrication for that domain.
  • Make the Voice specific. Name a distinct persona tone tied to the agent's expertise — never generic "friendly / professional / helpful." Two different agents must not share an interchangeable Voice line. Anchor it to a recognizable persona (an editor's red pen, a 3am-paged engineer, a cautious analyst) and a concrete manner (terse, warm, blunt, measured). If you couldn't tell this agent from another by its Voice alone, sharpen it.

Read the full file on GitHub · 78 lines

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.

  1. 2d ago First seen · 78 lines · 31 tokens per session scan A c6828d604408

Subscribe to this mod's changes

forge-agent is a command published in the GitHub repository emtcmca/promptsmith (2 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 31 tokens to every session and 1,127 once invoked, about $0.0002 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.