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 skills/langerrr/zforge/feature-executionnpx skills add Langerrr/zforge --skill feature-executiongit 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 | $0.00219 | $0.04545 |
| Opus 5 | $0.00110 | $0.02273 |
| Sonnet 5 | $0.00044 | $0.00909 |
| Haiku 4.5 | $0.00022 | $0.00455 |
Grade A, and why
feature-execution 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 — 254 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Feature Execution
The shared execution model behind $zforge:feature-orchestrate and $zforge:feature-resume. The two workflows differ only in who implements — a spawned agent or the planner in-session — and where decisions go — decision_review.md or the conversation. Everything below is the same for both.
Phase state
Phase state is read from the phase file's > Status: header and its contents. There is no signal protocol and no PID file.
| State | Meaning |
|---|---|
| PENDING | Not started |
| READY | Dependencies complete, not started |
| WAITING | Dependencies incomplete |
| RUNNING | An agent holds this phase now |
| REPORTED | An agent has reported DONE; the planner has not verified its evidence |
| COMPLETED | Accepted by the planner |
| PAUSED | Agent hit a pause trigger and stopped |
| FAILED | Unrecoverable error |
| INTERRUPTED | Agent died mid-phase — usage limit, API error, session end |
REPORTED is the state acceptance runs in. Nobody holds the phase and nothing has been accepted; the work is on disk and the verification is missing.
Spawning
One worker per phase. In Codex, spawn a subagent and instruct it to load $zforge:phase-agent; applicable skill instructions are allowed to request this delegation. In Claude Code, use the plugin's zforge:phase-agent agent type. Both routes use the same phase contract below.
The phase file is the authoritative prompt. The spawn message is a pointer to it plus the report contract — never a second copy of the phase's instructions. Two copies drift, and the copy the agent reads wins.
A spawn message contains only:
- The absolute path to the phase file, with the instruction to read it and its
## Required Contextbefore anything else. - A one-paragraph statement of the phase's postcondition and its required evidence classes.
- The report contract below.
- The budget-stop clause below.
Set the phase's > Status: to RUNNING before spawning.
The budget-stop clause
Every spawn message carries it:
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 · 254 lines · 219 tokens per session scan A 56a3d24047f8
feature-execution is a skill published in the GitHub repository Langerrr/zforge (10 stars, last pushed 2d ago), licensed MIT. It adds 219 tokens to every session and 4,545 once invoked, about $0.0011 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 skills, from other repositories
sdd-workflow
Guide a project with Spec-Driven Development (SDD) discipline - idea, approved spec, consistent plan, tasks, a gate that verifies approval and consent, implementation, validation, and logbook. Bilingual EN/ES. Use when the user wants to start, spec, plan, implement, or validate work with SDD, or mentions specs, plans…
eo-flow
单点 handoff：把 eo- 任务（review/test/implement/change-review/spec-review）甩给 tmux 里的 codex pane 执行，回包后决定"甩回去修"或"暂停问用户"。触发：eo-flow / 甩给 codex / /eo-flow。 NOT FOR: 完整流水线（用 /eo-workflow）。.
eo-project-init
所有 eo- skill 的总入口。其它 skill（eo-change / eo-implement / eo-doc-manager / …）都依赖 .eo-project.json；未运行过本 skill 的项目无法使用其它 eo- skill。.
eo-workflow
在 tmux 多窗格中编排 eo- 技能流水线（module-init / change / implement / archive），跨 pane 自动派发和状态轮询。触发：eo-workflow / 启动工作流 / 全流程 / /eo-workflow。.
eo-project-update
更新项目进度、阶段状态、决策记录，并同步项目看板（若配置）。通过 .eo-project.json 定位项目。触发：更新项目 / 同步进展 / 记录决策 / 阶段完成 / /eo-project-update。.
eo-backlog
往项目 backlog.md 追加待办、灵感或未接入的未来规划。通过 .eo-project.json 定位项目。触发：加入 backlog / 记一条待办 / 以后再说 / /eo-backlog。.