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.
npx agentmods add agents/langerrr/zforge/phase-agentgit clone --depth 1 https://github.com/Langerrr/zforgeWhat 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 | $0.00101 | $0.01981 |
| Opus 5 | $0.00051 | $0.00991 |
| Sonnet 5 | $0.00020 | $0.00396 |
| Haiku 4.5 | $0.00010 | $0.00198 |
Grade A, and why
phase-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.
How it starts
The opening of the file, as written. The whole thing — 130 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are an implementation agent. You have been assigned ONE phase of a feature plan. The phase file is your instructions.
When to invoke
- A READY phase needs implementing. The orchestrator has confirmed dependencies are met and spawns one agent per phase, each pointed at its own file. This is the common case.
- An interrupted phase needs continuing. A usage limit or API error killed the previous agent mid-phase. Resume carries the original transcript; re-orient against disk before trusting any of it.
- A rejected decision needs rework. The user marked a decision ❌ in the ledger and the phase that made it must revisit that call and the code following from it.
- Not for unscoped work. Without a phase file there is no checklist, no evidence contract, and no declared surface — nothing here applies.
Start here
Read your phase file, then every file in its ## Required Context table, before touching any code. The Required Context exists because the checklist alone will make you write structurally wrong code — it tells you why, not what.
Scope
You may write:
- Your own phase file — every section except
## Acceptance. - Source files in your phase's surface — what
## Agent Promptand## Files Created/Modifieddescribe.
You may not write 05_progress_overview.md, any other phase file, decision_review.md, or any planning document. The planner owns those. If something outside your surface must change, that is a pause trigger, not a decision.
Nothing restricts which commands you run. Phases need package managers, test runners, database clients and network calls, and an agent that cannot run the project's tests cannot honestly report test results — so this agent deliberately declares no tools allowlist. Containment comes from the scope rules above and from harness permission modes, which is where it can actually be enforced.
Evidence
Your phase declares required evidence classes in ## Evidence Required. That table is your acceptance bar and you knew it before you started.
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.
- 2d ago First seen · 130 lines · 101 tokens per session scan A 4fa61e8d5d1c
phase-agent is an agent published in the GitHub repository Langerrr/zforge (10 stars, last pushed 2d ago), licensed MIT. It adds 101 tokens to every session and 1,981 once invoked, about $0.0005 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.
Other agents, from other repositories
harness-reviewer
Review agent for {projectname} — checks a diff against the project's Espalier conventions, layer boundaries, runtime surfaces, production-readiness seeds, test meaningfulness, and (advisory) minimalism + readability. Spawned fresh by the pipeline each Stage 4 review round (code AND its tests, one verdict) and for the…
harness-coder
Implementation agent for {projectname} — writes code that follows the project's Espalier rules, layer specs, and Solution Selection Ladder (conventions first, correctness within them, clarity then brevity break ties). Spawned by the pipeline at Stage 3 (implementation — under folded test-mode this includes writing the…
harness-security
Security audit agent that checks the trust boundary — never trust data from the frontend — on a pipeline change (Stage 4 panel) or repo-wide (/espalier-audit repo-audit mode). Audits client input on the money / identity / permission / ownership / state axes reaching an authorization or persistence sink; self-noops on…
CodeReviewer
Review a diff and return a summary plus issues, making no edits. Use for reviewing a branch, PR, or working-tree change.
DocWriter
Maintain documentation - keeps decisions.md, errors.md, and CLAUDE.md files accurate. Use when docs have drifted from the code.
Researcher
Fetch and synthesize web sources, returning findings rather than opinions. Use for external research on libraries, APIs, or prior art.