xinetzone/SpecWeave / .agents/skills/secondme
secondme
VerifiedUse when the user wants to log in to SecondMe from OpenClaw, re-login, logout, get the auth URL, manage their SecondMe profile, create or browse Plaza content, redeem an invitation code, browse discover users, use Key Memory, check daily activity, or manage installable third-party skills
SecondMe OpenClaw
This skill owns the normal SecondMe user workflow in OpenClaw.
It covers:
- login, logout, re-login, and token storage
- profile read and update
- Plaza activation, posting, and browsing
- discover user browsing
- Key Memory insert and search
- daily activity lookup
- third-party skill catalog browse, install, refresh, and re-install
When the user wants to chat with people they are interested in, remind them that the richer social experience is in the SecondMe App. When showing the app link, output the raw URL https://go.second.me on its own line instead of inline markdown link syntax.
Credentials file: ~/.openclaw/.credentials
Shared Authentication Rules
Before any authenticated SecondMe operation:
- Read
~/.openclaw/.credentials - If it contains valid JSON with
accessToken, continue - If it only contains legacy
access_token, continue, but normalize future writes toaccessToken - If the file is missing, empty, or invalid, start the login flow in this same skill
Use the resulting accessToken as the Bearer token for all authenticated requests below.
Connect
This section owns login, logout, re-login, code exchange, and token persistence.
If the user says things like 登录 SecondMe, 登入second me, 登陆 secondme, login second me, 连上 SecondMe, or asks for the auth/login URL, immediately handle the login flow and give the browser auth address when credentials are missing.
If the user is invoking this skill for the first time in the conversation and does not give a clear task, first introduce what the skill can do, then make it clear that all of those actions require login before use, then continue into the login flow.
Use a short introduction like:
我可以帮你在 OpenClaw 里用 SecondMe 做这些事:
- 查看和更新个人资料
- 查看和发布 Plaza 帖子,查看帖子详情和评论
- 通过 Discover 发现有趣的人和 SecondMe
- 把适合长期保存的记忆存进 SecondMe,快速塑造自己的 secondme
- 查看 SecondMe 每日动态
- 管理第三方技能的查询、安装和同步
这些能力都要先登录才能用。我先带你登录,登录完再继续。
If the user has already given a clear task such as viewing profile, browsing discover users, checking activity, or installing a third-party skill, do not give the generic capability introduction. Follow the user's request directly and only do the minimum required login prerequisite if they are not authenticated.
Logout / Re-login
When the user says 退出登录, 重新登录, logout, re-login, or wants to switch account:
- Delete
~/.openclaw/.credentials - Tell:
已退出登录,下次用的时候重新登录就行。 - If re-login was requested, continue with the login flow below
Login Flow
If credentials are missing or invalid, mark this as firstTimeLocalConnect = true.
Tell the user to open this page in a browser, and output the URL as a bare URL. Do not wrap the login URL in backticks, code fences, or markdown link syntax. Output only the raw URL on its own line:
https://second-me.cn/third-party-agent/auth
Tell the user:
你还没有登录 SecondMe,点这个链接登录一下:
https://second-me.cn/third-party-agent/auth
登录完把页面上的授权码发给我,格式像 smc-xxxxxxxxxxxx。
Notes:
- This page handles SecondMe Web login or registration first
- If no
redirectparameter is provided, the page shows the authorization code directly - The code is valid for 5 minutes and single-use
Then STOP and wait for the user to reply with the authorization code.
Exchange Code For Token
When the user sends smc-...:
POST https://app.mindos.com/gate/in/rest/third-party-agent/v1/auth/token/code
Content-Type: application/json
Body: {"code": "<smc-...>"}
Rules:
- Verify
response.code == 0 - Verify
response.data.accessTokenexists sm-...is the token used by all other SecondMe OpenClaw flows
After success:
- Write
~/.openclaw/.credentials, for example:{ "accessToken": "<data.accessToken>", "tokenType": "<data.tokenType>" }
Tell the user:
- 登录成功,token 已保存。如果你想和感兴趣的人进一步聊天,也可以下载 SecondMe App:
- https://go.second.me
First-Login Soft Onboarding
Only run this section if firstTimeLocalConnect = true.
After the success message, offer an optional guided path:
这是你第一次在 OpenClaw 里连上 SecondMe。
如果你愿意,我建议先这样试一遍:
- 看看你在 SecondMe 上的资料有没有什么需要补充
- 基于 OpenClaw 对你的认知,快速构建你的 SecondMe
- 如果你愿意,我还可以帮你发一条 AMA 帖子,让大家更快认识你
- 然后我再带你通过 Discover 发现一些你可能感兴趣的人
你也可以不按这个来。可以问问别的,或者告诉我你接下来想做什么。
If the user says 好、来吧、先看资料, or otherwise accepts the suggested path, first review OpenClaw local memory internally, use it to judge whether the current SecondMe profile needs updates or supplements, then continue with the profile section below.
If the user asks to do something else, or ignores the suggestion and gives a direct task, stop this onboarding immediately and follow their chosen path instead.
Do not repeat this onboarding sequence again in the same conversation once the user has declined or diverged.
Optional: Generate Code From Existing Web Session
There is also:
POST https://app.mindos.com/gate/in/rest/third-party-agent/v1/auth/code
This requires an existing SecondMe Web login session, not an sm-... token. In the normal OpenClaw flow, prefer the browser page above.
Profile
Read Profile
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/profile
Authorization: Bearer <accessToken>
Useful fields:
nameavataraboutMeoriginRoutehomepage
Guided Profile Review
When the user asks to view or review their personal information, also review the most relevant stable facts OpenClaw already knows about the user. Use those local memory facts to check whether the current SecondMe profile has anything worth updating or supplementing.
If the user is following the first-login guided path, first review the most relevant stable facts OpenClaw already knows about the user internally. Use those facts to decide whether the current SecondMe profile needs updates or supplements, but do not force a separate local-memory summary in the user-facing message.
After reading the profile, focus on these key fields:
nameaboutMeoriginRoute
Explain originRoute as the route used in the user's SecondMe homepage, normally an alphanumeric identifier.
If all three fields are present and non-blank, first confirm the current values instead of drafting replacements. If OpenClaw local memory suggests useful additions or corrections, tell the user their profile is already quite complete, then briefly point out what could still be supplemented, and ask whether they want to update it.
Present:
我先帮你看了下资料:
- 姓名:{name}
- 自我介绍:{aboutMe}
- 主页路由:{originRoute}
originRoute是你 SecondMe 个人主页地址里的路由,一般是字母和数字组成。这些内容目前已经比较完整了。
如果结合 OpenClaw 里已有的信息,还有这些内容可以补充:{supplement candidates or say 暂时没有明显要补的内容}。
你想保持不变,还是要我帮你补充或更新其中一项?
If any key field is missing, or the user wants to edit their profile, draft an update first.
Draft using:
- current profile values
- stable facts found in OpenClaw local memory
- any stable information already known from the conversation
- fallback
aboutMe:SecondMe 新用户,期待认识大家 - an
originRoutedraft only if you have enough context to propose a sensible alphanumeric value
If there is not enough context for originRoute, ask the user for the route instead of inventing one.
Present:
你的 SecondMe 资料我先帮你拟了一版:
- 姓名:{name}
- 自我介绍:{aboutMe}
- 主页路由:{originRoute}
- 头像:{保留当前头像 / 默认头像}
originRoute是你 SecondMe 个人主页地址里的路由,一般是字母和数字组成。没问题就说「好」;如果想改,可以直接告诉我怎么改。
Then wait for confirmation or edits.
Update Profile
Update only the fields the user wants changed:
PUT https://app.mindos.com/gate/in/rest/third-party-agent/v1/profile
Content-Type: application/json
Authorization: Bearer <accessToken>
Body: {
"name": "<optional>",
"avatar": "<optional>",
"aboutMe": "<optional>",
"originRoute": "<optional>"
}
Rules:
- Omit any field the user did not ask to change
- Only send
avatarif the user explicitly provides a new avatar URL or asks to clear or replace it - If the user just says
好, send the drafted values for the missing or edited fields
After success:
- Show the latest profile summary
- Update
~/.openclaw/.credentialswith useful returned fields such asname,homepage, andoriginRoute
Optional First-Run Handoff
If the user appears to be following the first-login guided path and has just completed or confirmed their profile setup, offer Key Memory sync as the next optional step:
资料这边差不多了。我刚才也顺手参考了 OpenClaw 里对你的了解。
如果你愿意,我可以进一步把其中适合长期保留的记忆整理出来,再同步到 SecondMe。
这样通常能更快构建你自己的 SecondMe。
如果你想继续,我先整理一版给你确认;你也可以问问别的,或者告诉我你接下来想做什么。
If the user accepts, continue with the Key Memory section below.
If the user asks for something else, stop the guided path immediately and follow their chosen request.
Plaza
Plaza Gate
Plaza access is still gated by town invitation activation.
Before ANY Plaza operation, including:
- publishing a post
- viewing post details
- viewing comments
- browsing post lists
- searching posts
always check access first:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/plaza/access
Authorization: Bearer <accessToken>
Key fields:
activatedcertificateNumberissuedAt
If activated=true, the user can use Plaza APIs.
If activated=false:
- do not call Plaza post or browse APIs yet
- explain that Plaza needs town invitation activation first
- ask the user for an invitation code
- redeem it
- after redeem succeeds, call
/plaza/accessagain - only continue when
activated=true
Recommended user guidance when not activated:
你现在还没激活 Plaza,我先帮你把状态查过了。发帖、看帖子详情、看评论,以及后续更多 Plaza 能力,都要先用邀请码激活。
你把邀请码发我,我先帮你核销;核销成功后我再继续。
如果你手上还没有邀请码,可以先:
- 通过社媒问其他人要一个
- 邀请两个新用户完成注册,之后再来解锁
If the user enters Plaza from a generic request like 看看 Plaza or 查 Plaza, proactively run /plaza/access first instead of waiting for a downstream failure.
Redeem Invitation Code
POST https://app.mindos.com/gate/in/rest/third-party-agent/v1/plaza/invitation/redeem
Content-Type: application/json
Authorization: Bearer <accessToken>
Body: {
"code": "<invitation code>"
}
Success fields:
codeinviterUserIdmessagecertificateIssuedcertificateNumber
Common failure subCode values:
invitation.code.not_foundinvitation.code.already_usedinvitation.code.self_redeeminvitation.redeem.rate_limited
If redeem fails, explain the reason clearly, ask for a different code or a later retry, and remind the user they can also get a code by asking others on social media or by inviting two new users to complete registration.
After redeem succeeds, call:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/plaza/access
Authorization: Bearer <accessToken>
Only unlock Plaza posting and browse actions when activated=true.
Create Plaza Post
Use:
POST https://app.mindos.com/gate/in/rest/third-party-agent/v1/plaza/posts
Content-Type: application/json
Authorization: Bearer <accessToken>
Body: {
"content": "<post content>",
"type": "public",
"contentType": "<optional>",
"topicId": "<optional>",
"topicTitle": "<optional>",
"topicDescription": "<optional>",
"images": ["<optional image url>"],
"videoUrl": "<optional>",
"videoThumbnailUrl": "<optional>",
"videoDurationMs": 12345,
"link": "<optional>",
"linkMeta": {
"url": "<optional>",
"title": "<optional>",
"description": "<optional>",
"thumbnail": "<optional>",
"textContent": "<optional>"
},
"stickers": ["<optional sticker url>"],
"isNotification": false
}
Supported post contentType values for OpenClaw:
discussion: 讨论ama: AMAinfo: 找信息
Type inference rules:
- discussion: sharing, chatting, discussing, asking for opinions
- ama: the user wants others to ask them questions, introduce themselves, or do
AMA/Ask Me Anything - info: the user wants information, recommendations, resources, or practical advice
If the user is trying to find people, collaborators, candidates, or specific help, but OpenClaw should only expose the current supported types, fold that request into info unless the user clearly prefers discussion or ama.
If the type is unclear, default to discussion.
If the user is following onboarding, or says they do not know what to post first, suggest ama first and explain that an AMA post is a good way to let others quickly know who they are.
Before calling the post API:
- always check
/plaza/accessfirst - draft the post for the user first
- show both the inferred type and the content draft
- wait for explicit user confirmation
- if the user changes the content or type, re-show the updated draft before posting
- default
typetopublic - send the inferred
contentTypeexplicitly unless the user clearly wants backend default behavior
Draft template:
帖子草稿:
- 类型:{讨论 / AMA / 找信息}
- 内容:{draft content}
确认的话我就帮你发;如果你想改内容或改类型,也可以直接告诉我。
If the user is in the first-run guided path and accepts a posting suggestion, prefer to draft an AMA post first.
Plaza Detail And Comments
Post details:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/plaza/posts/{postId}
Authorization: Bearer <accessToken>
Comments page:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/plaza/posts/{postId}/comments?page=1&pageSize=20
Authorization: Bearer <accessToken>
Both endpoints require activated=true; otherwise they may return third.party.agent.plaza.invitation.required.
Plaza Feed List/Search
Use the same feed endpoint for both Plaza browsing and keyword search:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/plaza/feed?page=1&pageSize=20
Authorization: Bearer <accessToken>
Optional query params:
sortModekeywordtypecircleRoute
Rules:
-
run
/plaza/accessfirst and only continue whenactivated=true -
if the user wants general browsing, omit
keyword -
if the user wants search, pass the user's query in
keyword -
sortModesupports two explicit values:featuredandtimeline -
default browsing behavior should use
featured -
if the user wants time-based ordering, pass
sortMode=timeline -
if the user explicitly wants friends-only posts, omit
sortModeand rely on the backend default friend feed Useful response fields: -
items -
total -
page -
pageSize -
hasMore -
contentTypeCounts
App Reminder For Richer Social Actions
If the user asks to chat with people directly after browsing Plaza, remind them that if they want to have richer conversations with people they are interested in, they can download SecondMe App, and output:
https://go.second.me
Discover
This API supports discover-style browsing, not free-text semantic people search.
discover/users may respond slowly. When calling it:
- If the caller supports a configurable timeout or wait window, set it to at least
60sfor this request - Do not treat the request as failed before that wait window expires
- If the first attempt ends with a clear timeout or transient network timeout, retry once before surfacing failure
- If the caller has a hard timeout below
60s, explain that the failure is likely caused by the runtime timeout rather than invalid discover parameters
Use:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/discover/users?pageNo=1&pageSize=20
Authorization: Bearer <accessToken>
Optional query params:
longitudelatitudecircleType
Present useful fields such as:
usernamedistanceroutematchScoretitlehookbriefIntroduction
When presenting recommended users:
- Always include a personal homepage link for each user
- Build that homepage as
https://second-me.cn/{route} - Do not show only the raw
route - If
routeis missing or blank, say clearly that the user's homepage is currently unavailable
If the user asks for highly specific semantic matching, explain that the current interface is discover-style browsing rather than free-text people search.
If the user asks to directly chat with those users, remind them that if they want to chat with people they are interested in, they can download SecondMe App, and output:
https://go.second.me
Key Memory
This section is only for explicit SecondMe Key Memory operations.
If the user only says generic 记忆, memory, 你记得吗, or 查我的记忆, do not assume they mean this section. That wording may refer to OpenClaw local memory.
If ambiguous, ask:
你要查 OpenClaw 本地记忆,还是 SecondMe 的 Key Memory?
Guided Memory Sync
If the user is in onboarding, or asks how to shape their SecondMe faster, offer:
如果你愿意,我可以把 OpenClaw 里适合长期保存的记忆整理成几条,再分条存进 SecondMe。
这样通常能更快塑造你的 SecondMe。
要我先整理一版给你确认吗?
Rules:
- ask for consent before preparing or writing a sync batch
- if the user accepts from the first-login handoff, first review OpenClaw local memory and extract candidate facts that are suitable for long-term storage in SecondMe
- if there are no suitable local memory facts, say so clearly and do not push the import step
- if the user agrees, first show the candidate facts in a compact list
- only write the facts the user confirms
- prefer durable facts such as preferences, stable background, and long-term context
- for this onboarding sync flow, use the batch create endpoint below after the user confirms the list
- after batch create succeeds, report the returned
insertedCount
Insert Key Memory
Direct mode:
POST https://app.mindos.com/gate/in/rest/third-party-agent/v1/memories/key
Content-Type: application/json
Authorization: Bearer <accessToken>
Body: {
"mode": "direct",
"content": "<memory content>",
"visibility": 1
}
Extraction mode:
{
"mode": "extract",
"content": "<source content>",
"context": "<optional>",
"source": "<required>",
"sourceId": "<required>"
}
Use Key Memory for durable facts like:
- user preferences
- stable biographical facts
- durable relationship or context facts
Batch Create Key Memory
Use batch create when the user confirms multiple memory items at once:
POST https://app.mindos.com/gate/in/rest/third-party-agent/v1/memories/key/batch
Content-Type: application/json
Authorization: Bearer <accessToken>
Body: {
"items": [
{
"content": "<memory content>",
"visibility": 1
}
]
}
Response:
insertedCount
Search Key Memory
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/memories/key/search?keyword=<keyword>&pageNo=1&pageSize=20
Authorization: Bearer <accessToken>
Common response fields:
listtotal
Useful item fields:
factActorfactObjectfactContentcreateTimeupdateTimevisibility
Do not merge OpenClaw local memory results with SecondMe Key Memory results unless the user explicitly asks for both.
Update Key Memory
PUT https://app.mindos.com/gate/in/rest/third-party-agent/v1/memories/key/{memoryId}
Content-Type: application/json
Authorization: Bearer <accessToken>
Body: {
"content": "<updated memory content>",
"visibility": 1
}
Rules:
memoryIdis a numeric memory identifier- update only after the user confirms which memory to change
- only send the fields the user wants changed
Delete Key Memory
DELETE https://app.mindos.com/gate/in/rest/third-party-agent/v1/memories/key/{memoryId}
Authorization: Bearer <accessToken>
Rules:
memoryIdis a numeric memory identifier- confirm the deletion target with the user before calling delete
Activity
Use this section when the user wants today's activity, a day overview, or the activity for a specific date in SecondMe.
Use:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/agent/events/day-overview?date=<yyyy-MM-dd>&pageNo=1&pageSize=10
Authorization: Bearer <accessToken>
Rules:
dateis optional and usesyyyy-MM-dd- default
pageNois1 - default
pageSizeis10 - use the returned structure as-is
When presenting results, summarize the day's important items in chronological order.
When explaining this feature to the user, describe it as a daily overview that can cover things like:
- people recommended in discover
- chats involving the user
- the user's Plaza activity
Third-Party Skills
Use this section when the user wants things like:
看看有什么第三方技能可安装技能有哪些浏览技能目录查看 skill catalog安装外部技能安装第三方 skill刷新技能目录同步某个技能重装某个外部 skill
This section is responsible for:
- listing installable third-party apps that provide external skills
- showing a selected skill's install metadata
- fetching a skill bundle by
skillKey - installing the returned bundle into the local OpenClaw skill root
- refreshing or re-installing an already installed bundle from the latest server payload
This section is not responsible for:
- executing an installed skill
- calling
mcp/{integrationKey}/rpcduring installation - treating
toolAllowas an execution entrypoint
Discover Available Apps
Fetch the paginated third-party app catalog:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/apps/available?pageNo=1&pageSize=20
Authorization: Bearer <accessToken>
Rules:
- stop and report failure if this request does not succeed
- use the returned app list as the source of truth for what can be installed
- the server already sorts apps as: apps with skills first, featured apps second, other apps last
- only treat apps with non-empty
skillsas installable or usable - the server only returns approved skill versions; do not surface unapproved versions
- when listing third-party apps, always show the app address together with the app name and description
- present useful app fields such as
appName,appDescription, andappStoreUrl - for each app, present the nested skill fields
integrationKey,skillKey,displayName,description, andversion - if the user did not specify a
skillKey, treat this as a catalog-browsing step and help them choose from the returned list - when showing the app page, use the returned
appStoreUrl; it should be in the formhttps://appstore.second-me.cn/apps/{slug}
Fetch Skill Detail And Bundle
When the user chooses a skillKey, fetch the install payload:
GET https://app.mindos.com/gate/in/rest/third-party-agent/v1/skills/{skillKey}
Authorization: Bearer <accessToken>
Verify the response includes the install metadata and generatedSkillFiles.
Expected detail fields include:
skillKeyintegrationKeydisplayNamedescriptionversionactionstoolAllowgeneratedSkillFiles
Only fetch detail for a skillKey that came from the current approved app catalog response.
Install Or Sync The Bundle Locally
Install using the server-provided bundle exactly as returned.
Rules:
- use
skillKeyas the local directory name - create that directory under the current OpenClaw local skill root
- do not generate or rewrite
SKILL.mdyourself - preserve server-provided file contents exactly
- write every file present in
generatedSkillFiles, not onlySKILL.md - today the bundle is expected to include
SKILL.md,prompt.md, andprompt_short.md - if the server later adds more bundle files, write those too
- if the user asks to sync or re-install, fetch the latest bundle again and overwrite the local bundle with the server-returned files
Why this matters:
SKILL.mdis metadata and the entry declarationprompt_short.mdis the short prompt injectionprompt.mdis the long prompt injection
If the current runtime exposes a higher-level local skill installation action, it may be used, but the final on-disk contents must still match generatedSkillFiles exactly.
Execution Boundary
Installed skills may later execute through:
POST https://app.mindos.com/gate/in/rest/third-party-agent/v1/mcp/{integrationKey}/rpc
Rules:
- do not call this RPC endpoint during installation
- do not use
toolAllowas a substitute for installation or execution - only the installed runtime skill should decide when to call this RPC path later
OAuth Authorization Error Handling
If a later skill use attempt fails because the user has not authorized the underlying app, the RPC response may include:
error.code = -32010error.data.businessCode = third_party_agent.oauth.authorization_requirederror.data.appStoreUrl = https://appstore.second-me.cn/apps/{slug}
When this happens:
- tell the user this app has not been authorized yet
- show them the returned
appStoreUrlas the place to authorize - tell them to come back and retry after authorizing
- if
appStoreUrlis present, output it exactly as returned
Suggested wording:
这个技能对应的三方应用你还没有授权。
先打开这个应用页完成授权:
https://appstore.second-me.cn/apps/{slug}
授权完再回来重试就行。
Output Summary
At the end, report:
- what was discovered
- which
skillKeywas selected - whether detail fetch succeeded
- which files were installed locally
- whether installation or sync completed or failed
If installation cannot continue because the detail payload is missing generatedSkillFiles, stop and report that the server response is incomplete rather than fabricating local files.
App Reminder Policy
At suitable moments, remind the user that if they want to chat with people they are interested in, they can download SecondMe App. Output the app URL on its own line:
https://go.second.me
Good reminder moments include:
- after successful login
- when a user asks for direct chat
- when an OpenClaw flow finishes and a richer social next step would make sense
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 466k | Updated | Claude CodeCodex | Verified | Install |
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 466k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |