Agent skill

Comet Verify

by rpamis in rpamis/comet

A skill your agent uses when Comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾。

MITAuto-check passedAgent Workflows

Install Comet Verify

skills CLI
$ npx skills add rpamis/comet --skill comet-verify -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install rpamis/comet comet-verify --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/rpamis/comet.git skills-src && mkdir -p .claude/skills && cp -r skills-src/eval/local/skills/benchmarks/040-beta/comet-verify .claude/skills/comet-verify && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
comet-verify
GitHub stars
3.2k
Token cost
~1.9k tokens
SKILL.md length
591 words
Files
1
Skills in repo
52
Repo updated
First seen
Licence
MIT

At a glance

A skill your agent uses when Comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾。

  • Works in 4 steps: 改动规模评估 → 产物上下文加载(Hash 按需读) → 收尾(Superpowers) → …
  • Comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾
  • SKILL.md covers 前置条件, 步骤, 退出条件 and 上下文压缩恢复, plus 1 more section
  • Calls node, git and npm

What it does

Comet Verify is an agent skill from rpamis/comet. Use when Comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾。

Its SKILL.md is about 1.9k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Agent Workflows. The repository describes itself as: Comet: agent skill harness for turning ideas into evaluated workflows. The licence is MIT.

When your agent uses it

  • Comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾

Example prompts

  • “/comet-verify”

Workflow steps

4 steps, taken from the step headings in SKILL.md.

  1. 改动规模评估
  2. 产物上下文加载(Hash 按需读)
  3. 收尾(Superpowers)
  4. 记录验证证据

What it can do on your machine

Read from SKILL.md and the folder at commit 0fd42a0. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • node
    • git
    • npm
    • mvn
    • cargo

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use git and npm, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Comet Verify loads about 1.9k tokens when it runs. Until then it costs about 18 tokens; SKILL.md has 591 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~18
When it runs · the whole SKILL.md, loaded when a task matches
~1.9k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from rpamis/comet at commit 0fd42a0, republished under its MIT licence (© rpamis). 591 words, ~1,906 tokens.

Download SKILL.mdSave it as .claude/skills/comet-verify/SKILL.md (or your agent's skills folder).
name
comet-verify
description
Use when Comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾。

Comet 阶段 4:验证与收尾(Verify)

前置条件

  • 代码已提交(阶段 3 完成)
  • tasks.md 全部任务已完成

步骤

0a. 输出语言约束

验证报告和分支处理说明必须使用触发本次工作流的用户请求语言。

0b. 入口状态验证(Entry Check)

按 comet/reference/scripts.md 定位脚本(定位 comet-env.mjs),然后执行入口验证;从任意入口恢复时先按 comet/reference/context-recovery.md 运行恢复检查:

bash
node "$COMET_STATE" check <change-name> verify

验证通过后继续 Step 1。验证失败时脚本会输出具体失败原因。

幂等性:verify 阶段所有检查可安全重复执行。如 verify_result 已为 pass 且 branch_status 已为 handled,说明验证已完成,直接执行 guard 流转。如 verify_result 为 pending,从头开始验证。

1. 改动规模评估

执行规模评估:

bash
node "$COMET_STATE" scale <change-name>

脚本自动统计任务数、增量规格数、变更文件数,判断使用 light 或 full 验证模式,并设置 verify_mode 字段。判定规则(满足任一即 full):任务数 > 3、delta spec 能力数 > 1、变更文件数 > 8。

验证开始前,按 comet/reference/dirty-worktree.md 协议检查并处理未提交改动。verify 阶段的特殊处理:

  1. 若 dirty diff 属于当前 change 且涉及实现、测试、tasks、delta spec 或 design doc 变更,不在 verify 阶段直接修复或提交;报告失败项并进入 Step 1b 的验证失败决策阻塞点
  2. 若 dirty diff 只是 verify 本阶段产物(例如验证报告草稿、分支处理记录),可继续在 verify 阶段完成并记录状态
  3. 若 dirty diff 已实现但 tasks.md 未勾选,视为 build 状态滞后;报告失败项并进入 Step 1b,由用户决定回退修复或接受偏差

用户选择修复后,才允许回退到 build 阶段:

bash
# 仅在用户确认修复后执行
node "$COMET_STATE" transition <change-name> verify-fail

注意:verify-fail 回退到 build 时 branch_status 不会被重置。如果首次 verify 已完成分支处理,修复后再次进入 verify 时跳过已完成的分支处理步骤,直接使用 node "$COMET_STATE" set <change-name> branch_status handled 保留原有分支处理结果。

注意:如果 build 阶段每个任务都已提交,脚本基于工作区 diff 的文件数可能低估改动规模。此时必须读取 plan 文件头的 base-ref 并用提交区间复核:

bash
PLAN=$(node "$COMET_STATE" get <change-name> plan)
BASE_REF=$(grep '^base-ref:' "$PLAN" 2>/dev/null | head -1 | sed 's/^base-ref: *//')
git diff --stat "$BASE_REF"...HEAD

若提交区间显示改动超过轻量阈值(> 8 个文件、跨模块协调、或 delta spec 超过 1 个 capability),手动设置为完整验证:

bash
node "$COMET_STATE" set <change-name> verify_mode full

覆盖机制:如 agent 或用户认为自动评估结果不合适,可随时通过 node "$COMET_STATE" set <change-name> verify_mode <light|full> 手动覆盖。

1b. 验证失败决策(阻塞点)

验证不通过时必须按 comet/reference/decision-point.md 的协议暂停并等待用户决定修复或接受偏差。不得自动运行 node "$COMET_STATE" transition <change-name> verify-fail,也不得自动调用 /comet-build。

暂停时必须列出:

  • 失败项
  • 是否属于 CRITICAL 或 IMPORTANT(构建失败、测试失败、安全问题、核心验收场景失败、简化代码审查发现的正确性/安全/边界问题)
  • 推荐处理方式

不确定性原则:无法确定严重程度时,降级处理(SUGGESTION > WARNING > CRITICAL)。仅对构建失败、测试失败、安全问题使用 CRITICAL;模糊或不确定的问题标为 WARNING 或 SUGGESTION。

用户选择后按以下方式继续:

  • 全部修复:运行 node "$COMET_STATE" transition <change-name> verify-fail,然后调用 /comet-build 修复
  • 逐项处理:CRITICAL 或 IMPORTANT 失败项必须修复;WARNING/SUGGESTION 失败项可选择接受偏差,但必须在验证报告中记录接受原因和影响范围。若存在任何 CRITICAL 或 IMPORTANT 失败项,不允许跳过修复直接全部接受

重试上限:连续 3 次 verify-fail 循环后,第 4 次失败时代理不得自动选择继续修复;必须使用当前平台可用的用户输入/确认机制暂停,仅给出两个选项:「接受所有偏差并记录」或「继续修复」,由用户明确决定。

2. 产物上下文加载(Hash 按需读)

验证需要读取 OpenSpec 产物时,先检查产物是否自 design 阶段以来发生变化:

bash
RECORDED_HASH=$(node "$COMET_STATE" get <change-name> handoff_hash)
CURRENT_HASH=$(node "$COMET_HANDOFF" <change-name> --hash-only 2>/dev/null || echo "")
  • 若 RECORDED_HASH = CURRENT_HASH 且均非空且均非 null:OpenSpec 产物未变化,tasks.md 无需重新读取全文(用 grep -c '\- \[ \]' tasks.md 确认完成数即可)。proposal.md、design.md、delta spec 仍需读取用于对照检查。
  • 若 RECORDED_HASH 为空、为 null、或与 CURRENT_HASH 不一致:产物已变化或 hash 未记录,正常读取所有所需文件全文。

此优化仅跳过 tasks.md 的重复全文读取。proposal.md 和 design.md 包含验证检查项所需的完整上下文,不得因 hash 匹配而跳过。

立即执行: 使用 Skill 工具加载 Superpowers verification-before-completion 技能。禁止跳过此步骤。

技能加载后,按 verify_mode 分支执行:

2a. 轻量验证(小改动)

按以下 6 项进行检查:

  1. tasks.md 全部任务已完成 [x]
  2. 改动文件与 tasks.md 描述一致(git diff --stat / git diff --cached --stat / git diff --stat <base-ref>...HEAD 对照 tasks 内容)
  3. 编译通过(执行项目对应的构建命令,如 npm run build、mvn compile、cargo build 等)
  4. 相关测试通过
  5. 无明显安全问题(无硬编码密钥、无新增 unsafe 操作)
  6. 代码审查策略:当 review_mode: standard 或 thorough 时,必须使用 Skill 工具加载 Superpowers requesting-code-review 技能,请求只检查正确性、安全、边界条件的轻量代码审查;当 review_mode: off 时跳过自动代码审查,并在验证报告中记录跳过原因

简化代码审查的输入应限定为本次改动 diff、tasks.md 和必要的测试结果;审查范围只覆盖实现正确性、安全风险和边界条件,不执行 spec 覆盖率、Design Doc 一致性或漂移检查。若审查发现 CRITICAL 或 IMPORTANT 问题,按验证失败处理并进入 Step 1b。review_mode: off 只跳过自动 code review,不跳过构建、测试、安全检查或异常调试协议。

与 build 阶段审查的去重:若 build 阶段(executing-plans 或 subagent-driven-development)已按 review_mode 对同一 diff 完成最终代码审查,verify 的这次轻量审查聚焦「实现是否符合 spec/tasks 的正确性」与「build 之后新增的改动」,不重复评审 build 已审过且未变化的 diff。

通过标准:6 项全部 OK,无 CRITICAL 或 IMPORTANT 问题。

不通过时:报告失败项,进入 Step 1b 的验证失败决策阻塞点。用户选择修复后,才执行以下命令记录失败并回退到 build 阶段,然后调用 /comet-build 修复:

bash
# 仅在用户确认修复后执行
node "$COMET_STATE" transition <change-name> verify-fail

报告格式:简表列出 6 项检查结果 + PASS/FAIL。

跳过项(不在轻量验证中检查):

  • spec scenario 覆盖率
  • design doc 一致性深度比对
  • 不影响正确性、安全、边界条件的 code pattern consistency 建议
  • delta spec 与 design doc 漂移检测
Show full SKILL.md (223 more words)Show less
2b. 完整验证(大改动)

当规模评估结果为"大"时:

立即执行: 使用 Skill 工具加载 openspec-verify-change 技能。禁止跳过此步骤。

技能加载后,按其指引验证。检查项:

  1. tasks.md 全部任务已完成([x])
  2. 实现符合 openspec/changes/<name>/design.md 高层设计决策
  3. 实现符合 Design Doc(docs/superpowers/specs/ 下的技术设计文档)
  4. 能力规格场景全部通过
  5. proposal.md 目标已满足
  6. delta spec 与 design doc 无矛盾(若 Build 阶段有增量修改 spec,检查 design doc 是否有对应记录)
  7. docs/superpowers/specs/ 关联的设计文档可定位(文件存在且与当前 change 相关)

验证不通过时:报告缺失项,进入 Step 1b 的验证失败决策阻塞点。用户选择修复后,才执行以下命令记录失败并回退到 build 阶段,然后调用 /comet-build 补充:

bash
# 仅在用户确认修复后执行
node "$COMET_STATE" transition <change-name> verify-fail

Spec 漂移处理(用户决策点):

  • 若检查项 6 发现矛盾(delta spec 有内容但 design doc 未体现),必须使用当前平台可用的用户输入/确认机制以单选题形式暂停并等待用户选择处理方式,不得自动选择。选项:
    • 选项 A:在 design doc 追加 "Implementation Divergence" 节记录偏差原因。选项 A 属于 verify 阶段允许产物;写入后不得因该 design doc 变更再次触发 Step 1b dirty-worktree 决策
    • 选项 B:用户选择 B 后,运行 node "$COMET_STATE" transition <change-name> verify-fail,然后调用 /comet-build;由 /comet-build 的 Spec 增量更新规则加载 Superpowers brainstorming 更新 Design Doc + delta spec
    • 选项 C:确认偏差可接受,继续验证(归档时 design doc 将标记为 superseded-by-main-spec)
3. 收尾(Superpowers)

立即执行: 使用 Skill 工具加载 Superpowers finishing-a-development-branch 技能。禁止跳过此步骤。

如 Superpowers finishing-a-development-branch 技能不可用,停止流程并提示安装或启用 Superpowers 技能,不要用普通对话替代该步骤。

技能加载后,按其指引收尾。分支处理选项:

  1. 本地合并到主分支
  2. 推送并创建 PR
  3. 保持分支(稍后处理)
  4. 丢弃工作

这是用户决策点。必须按 comet/reference/decision-point.md 的协议暂停并等待用户选择分支处理方式,不得根据推荐、默认值或当前分支状态自行选择。只有在用户完成选择且对应操作完成后,才允许写入 branch_status: handled。

确认项:

  • 全部测试通过
  • 无硬编码密钥或安全问题
4. 记录验证证据

验证报告必须落盘,并在 .comet.yaml 中记录;分支处理完成后也必须写入状态字段。不要手动设置 verify_result: pass,由阶段守卫 --apply 推进。

bash
mkdir -p docs/superpowers/reports
# 将本次验证结论写入报告文件,例如:
# docs/superpowers/reports/YYYY-MM-DD-<change-name>-verify.md

node "$COMET_STATE" set <change-name> verification_report docs/superpowers/reports/YYYY-MM-DD-<change-name>-verify.md
node "$COMET_STATE" set <change-name> branch_status handled

退出条件

  • 验证报告通过
  • 分支已处理
  • .comet.yaml 中 verification_report 指向已存在的验证报告文件
  • .comet.yaml 中 branch_status: handled
  • 阶段守卫:运行 node "$COMET_GUARD" <change-name> verify --apply,全部 PASS 后由守卫通过 comet-state transition verify-pass 推进到 phase: archive(此步骤更新 phase 字段,与 auto_transition 无关)

验证和分支处理均完成后,运行阶段守卫推进 phase(此步骤与 auto_transition 无关):

bash
node "$COMET_GUARD" <change-name> verify --apply

状态文件自动更新为 phase: archive、verify_result: pass、verified_at: YYYY-MM-DD。

上下文压缩恢复

按 comet/reference/context-recovery.md 执行,phase 参数为 verify。

自动衔接下一阶段

按 comet/reference/auto-transition.md 执行。关键命令:

bash
node "$COMET_STATE" next <change-name>
  • NEXT: auto → 调用 SKILL 指向的 skill 进入下一阶段
  • NEXT: manual → 不要调用下一 skill,按 HINT 提示用户手动运行 /<SKILL>
  • NEXT: done → 流程已完成,无需继续

注意:无论 NEXT 为 auto 还是 manual,comet-archive 进入后必须先执行归档前最终确认阻塞点,等待用户明确选择「确认归档」后才允许运行归档脚本。不得因为验证已通过就自动归档。

© rpamis, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in eval/local/skills/benchmarks/040-beta/comet-verify of rpamis/comet.

Open the folder on GitHubat commit 0fd42a0

Compare with similar skills

Comet Verify next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Comet Verify compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Comet Verify this skillrpamis/comet3.2k—~1.9kAutomated safety check: PassMIT
MCP Server Builderanthropics/skills180k63 repos~2.3kAutomated safety check: PassApache-2.0
Hook Development for Claude Code Pluginsanthropics/claude-plugins-official38k10 repos~4.1kAutomated safety check: NotesApache-2.0
Using Superpowersfarm-fe/farm5.6k35 repos~1.4kAutomated safety check: PassMIT
Executing Plans Inlineobra/superpowers297k2 repos~5.1kAutomated safety check: PassMIT
Skill CreatorAzure/azqr79589 repos~8.2kAutomated safety check: PassApache-2.0

Similar skills

  • MCP Server Builder

    anthropics/skills

    Official

    Guides the design and implementation of Model Context Protocol servers in TypeScript or Python, from tool naming and error messages to evaluation.

    180k GitHub starsUsed in 63 repos~2.3k tokens
    Agent WorkflowsAuto-check passed
  • Hook Development for Claude Code Plugins

    anthropics/claude-plugins-official

    Official

    Explains how to write Claude Code plugin hooks, both prompt-based checks and bash commands, for events such as PreToolUse, Stop and SessionStart.

    38k GitHub starsUsed in 10 repos~4.1k tokens
    Agent WorkflowsAuto-check: notes
  • Using Superpowers

    farm-fe/farm

    A skill your agent uses when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions

    5.6k GitHub starsUsed in 35 repos~1.4k tokens
    Agent WorkflowsAuto-check passed
  • Executing Plans Inline

    obra/superpowers

    Has the agent carry out an implementation plan itself, task by task in the current session, keeping a ledger, proving each step with a test and ending with one whole-branch review.

    297k GitHub starsUsed in 2 repos~5.1k tokens
    Agent WorkflowsAuto-check passed
  • Skill Creator

    Azure/azqr

    Official

    Create new skills, modify and improve existing skills, and measure skill performance.

    795 GitHub starsUsed in 89 repos~8.2k tokens
    Agent WorkflowsAuto-check passed
  • Claude Code Agent Development

    anthropics/claude-plugins-official

    Official

    Explains how to write agents for Claude Code plugins: the markdown file with YAML frontmatter, trigger descriptions, model and color settings, and system prompt design.

    38k GitHub starsUsed in 7 repos~2.8k tokens
    Agent WorkflowsAuto-check passed

More from rpamis/comet

All 52 skills in this repo
  • Comet

    rpamis/comet

    A skill your agent uses when 用户要启动或恢复 Comet 工作流,需要根据 active change、.comet.yaml、hotfix/tweak 意图路由到对应阶段 Skill。

    3.2k GitHub stars~2.7k tokensUpdated today
    Auto-check passed
  • Comet

    rpamis/comet

    Comet workflow entry. An agent skill from rpamis/comet.

    3.2k GitHub stars~686 tokensUpdated today
    Auto-check passed
  • Comet

    rpamis/comet

    Comet — OpenSpec + Superpowers dual-star development workflow.

    3.2k GitHub stars~4.4k tokensUpdated today
    Auto-check passed
  • Comet Archive

    rpamis/comet

    Archive and deliver a Classic change. An agent skill from rpamis/comet.

    3.2k GitHub stars~3.9k tokensUpdated today
    Auto-check passed
  • Comet Classic

    rpamis/comet

    Comet Classic workflow entry. An agent skill from rpamis/comet.

    3.2k GitHub stars~3k tokensUpdated today
    Auto-check passed
  • Comet Design

    rpamis/comet

    Complete the Classic technical design and obtain user confirmation.

    3.2k GitHub stars~4.3k tokensUpdated today
    Auto-check passed

Categories

Questions about Comet Verify

What does Comet Verify do?

A skill your agent uses when Comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾。. Comet Verify is an agent skill from rpamis/comet.

When should I use Comet Verify?

Comet Verify fits situations like: comet change 已完成 build 阶段,需要验证实现、处理验证失败决策或完成分支收尾.

How do I install Comet Verify in Claude Code?

Run `npx skills add rpamis/comet --skill comet-verify -a claude-code`. Or copy the skill folder (eval/local/skills/benchmarks/040-beta/comet-verify in rpamis/comet) into .claude/skills/comet-verify in your project. Claude Code loads it when a task matches its description.

How do I install Comet Verify in Codex?

Run `npx skills add rpamis/comet --skill comet-verify -a codex`. Or copy the skill folder (eval/local/skills/benchmarks/040-beta/comet-verify in rpamis/comet) into .agents/skills/comet-verify in your project. Codex loads it when a task matches its description.

Can I use Comet Verify in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add rpamis/comet --skill comet-verify -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/comet-verify, .gemini/skills/comet-verify, .github/skills/comet-verify and .opencode/skills/comet-verify in your project.

What does Comet Verify need to run?

Going by SKILL.md and its folder, Comet Verify needs the command-line tools its instructions call (node, git, npm, mvn and cargo).

Does Comet Verify access the network?

SKILL.md contains no URLs. Its commands use git and npm, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Comet Verify safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Comet Verify use?

Comet Verify is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Comet Verify use?

About 1.9k tokens (SKILL.md is roughly 7.6k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Comet Verify?

Skills that share tags, products or a category with Comet Verify: MCP Server Builder (anthropics/skills, 180k stars), Hook Development for Claude Code Plugins (anthropics/claude-plugins-official, 38k stars), Using Superpowers (farm-fe/farm, 5.6k stars) and Executing Plans Inline (obra/superpowers, 297k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Comet Verify?

rpamis (a GitHub organization) maintains it in rpamis/comet, which has 3,166 GitHub stars. The repository holds 52 skills in this directory. The repository was last updated on October 9, 2026.

Source: rpamis/comet on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.