一键在 Manus 中运行任何 Skill

cicd

星标5

分支2

更新时间2026年6月2日 21:19

colleague's CI/CD lane, layered on `devex pr` (the same PR-lifecycle CLI as `agex`, invoked under the `devex` name). Delegates lint / open / read / reply / delta to devex; adds two extensions — `status` (SonarCloud quality gate + hotspots + unresolved-thread tally) and `await` (read --wait + status with non-zero exit on Sonar ERROR or unresolved threads). Use when: creating PRs in colleague, handling review feedback, polling CI status, or the user says "create PR", "review comments", "address feedback", "resolve threads". Renamed from `pr-review` in steward 0.7.0; rebased on agex/devex in 0.12.0.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

agentculture

agentculture/colleague

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

相关职业SOC

基于 SOC 职业分类

软件开发工程师计算机与数学类职业·SOC 15-1252

文件资源管理器

6 个文件

SKILL.md

readonly

同仓库更多 Skills

同仓库

summarize-delivery

agentculture/colleague

Close the loop after an assign-to-workforce run by turning what actually happened into an accountability artifact — planned versus actual delivery, mid-work decisions, plan drift, evidence-backed delivery claims, and remaining work. The plan the user confirmed is the contract; this skill records where execution obeyed it, where it changed, and what is genuinely safe to claim as delivered. Runs on complete, partial, AND failed runs — failure is reported faithfully, never smoothed over. Use when the user says "summarize delivery", "delivery summary", "wrap up", "close the loop", "what did we actually ship", or "plan versus actual", or after assign-to-workforce merges (or fails to merge) a plan's waves. Authored and maintained in agentculture/devague (origin = devague); guildmaster pulls this skill from here and broadcasts it to the AgentCulture mesh — it is NOT vendored from guildmaster like the inbound skills here.

2026-07-105

ask-colleague

agentculture/colleague

Ask colleague — a *different* backend/model than you (e.g. a local vLLM Qwen) — to take a scoped repo task off your plate, then fold its answer back. The point isn't a stronger model; it's a second, independent mind, and that diversity is the value: `ask-colleague review` gets a candid second opinion on a diff, `ask-colleague explore` gets a fresh read of an area, `ask-colleague write` hands off a small implementation, `ask-colleague feedback` grades a finished work item (the ROI loop), and `ask-colleague clean` reaps stale/corrupt `colleague/*` branches a crashed run left behind (which can break `git fetch`). Pilot a running work item with `monitor`/`guide`/`stop`. Reach for it REFLEXIVELY, the way you'd lean over to the teammate at the next desk — not only when asked: before you present or open a PR on a non-trivial committed diff, run `review` for a diverse second opinion; for a fresh read of an unfamiliar area whose answer is independent of your current context, run `explore`. Both are read-only — isolate

2026-07-095

recall

agentculture/colleague

Search the shared eidetic memory store and get back ranked, provenanced records. Drives `eidetic recall` with four search modes — exact (verbatim substring), approximate (vector/semantic), keyword (BM25 lexical), and hybrid (a weighted blend of vector+keyword, the default) — each hit carrying its text, full metadata, a relevance `score`, and a freshness `signal`. Recall passively reinforces matched records (bumps last_recall + recall_count). Shadowed and archived records are excluded by default; use --include-shadowed / --include-archived to retrieve them. The store uses visibility-aware routing: PUBLIC records inside a git repo go to <repo-root>/.eidetic/memory (committed, team-shared); PRIVATE records, or any record outside a git repo, go to $HOME/.eidetic/memory (never committed). An explicit EIDETIC_DATA_DIR wins and short-circuits to that single dir. Recall reads both stores and merges. The wrapper defaults queries to this agent's PERSONAL, PUBLIC scope (`--scope colleague --visibility public`, suffix re

2026-06-265

remember

agentculture/colleague

Ingest records into the shared eidetic memory store so they can be recalled later. Drives `eidetic remember`: accepts one record as a JSON object, or a batch as NDJSON on stdin for bulk ingest. Upsert is idempotent by id (and dedups by content hash) — re-remembering updates in place, never duplicates. Stamps a `created` date on every record at ingest time. Accepts `supersedes` (id of the record this one replaces, for within-scope shadowing via `sweep`) and `links` (list of related-memory ids). The store uses visibility-aware routing: PUBLIC records inside a git repo go to <repo-root>/.eidetic/memory (committed, team-shared); PRIVATE records, or any record outside a git repo, go to $HOME/.eidetic/memory (never committed). An explicit EIDETIC_DATA_DIR wins and short-circuits to that single dir. The wrapper defaults records to this agent's PERSONAL, PUBLIC scope (`--scope colleague --visibility public`, suffix read from culture.yaml) so a plain /remember commits to <repo-root>/.eidetic/memory (in-repo, team-shar

2026-06-265

promote

agentculture/colleague

Graduate colleague from a task runner into a persistent Culture mesh peer — the born → trained → resident lifecycle transition. Mints + self-registers a stable mesh identity, selects channels, and (with --serve) connects to IRC and runs the resident until interrupted. Requires the opt-in [culture] extra. Use when an operator wants to elevate their colleague instance into a long-lived Culture member that owns a channel and answers messages directly.

2026-06-125

communicate

agentculture/colleague

Cross-repo + mesh communication from colleague: file tracked GitHub issues on sibling repos, fetch issues from sibling repos to inline current state into briefs, and send live messages to Culture mesh channels. Use when the next step lives outside colleague (a brief for a sibling-repo agent, a status ping for a Culture channel, or pulling an issue body + comments into context). Issue posts auto-sign with `- colleague (Claude)`; mesh messages are unsigned (the IRC nick is the speaker). Not for in-colleague issues — use `gh issue create` or the `cicd` skill for those. Renamed from `coordinate` in steward 0.8.0; absorbed `gh-issues` in 0.9.1. Issue I/O is backed by `agtag` (>=0.1) starting in 0.11.0.

2026-06-025

name	cicd
type	command
description	colleague's CI/CD lane, layered on `devex pr` (the same PR-lifecycle CLI as `agex`, invoked under the `devex` name). Delegates lint / open / read / reply / delta to devex; adds two extensions — `status` (SonarCloud quality gate + hotspots + unresolved-thread tally) and `await` (read --wait + status with non-zero exit on Sonar ERROR or unresolved threads). Use when: creating PRs in colleague, handling review feedback, polling CI status, or the user says "create PR", "review comments", "address feedback", "resolve threads". Renamed from `pr-review` in steward 0.7.0; rebased on agex/devex in 0.12.0.

CI/CD — colleague edition

This skill drives devex pr — the agentculture PR-lifecycle CLI, invoked under the devex name. devex is the same tool as agex (upstream agentculture/agex-cli); colleague standardizes on the devex name and never shells out to gh directly to open a PR.

devex pr provides the five core PR-lifecycle verbs — lint, open, read, reply, delta. Steward used to vendor parallel scripts for each; in 0.12.0 those vendored copies were dropped in favor of delegating to the CLI. What's left in this skill is the steward-specific gating layer:

status — SonarCloud quality gate, OPEN issues, hotspots, deploy preview URL, unresolved-inline-thread tally.
await — composes devex pr read --wait with status and gates on Sonar ERROR / unresolved threads. The single command to run after pushing a fix when you want "wake me when this PR is triage-able."

Those two are the steward unique surface today. The await combo verb landed natively in the CLI (agex-cli#41, now closed); the gate extras that aren't yet native — SonarCloud hotspots, deploy-preview URL, an explicit resolved/unresolved thread tally — are tracked upstream in agex-cli#52 and migrate out of this skill once they land.

The workflow is encapsulated in scripts/workflow.sh — follow that (or call devex pr directly).

The agex-cli inversion (upstream-as-consumer)

One consumer is special: agex-cli itself, the repo that owns the pr CLI (agex/devex are the same tool). Vendoring this skill there verbatim would re-vendor bash that just wraps the Python CLI it already ships, so agex-cli vendors it adapted-thin (agex-cli#53): workflow.sh is the only script and it forwards lint | open | read | reply | delta | await straight to the native pr <verb> — including the native pr await combo verb (agex-cli 0.21.0). The steward status / await shell extensions and the vendored helpers (pr-reply.sh, _resolve-nick.sh, portability-lint.sh) are all redundant there, each superseded by a native verb. For that one consumer the skill collapses to a pure delegate.

The only gate bits not yet native are SonarCloud hotspots, the deploy-preview URL, and an explicit resolved/unresolved thread tally — tracked upstream in agex-cli#52. Once those land, steward retires pr-status.sh too and workflow.sh status/await delegates to the native CLI everywhere.

For broadcasts: a skill-update brief to agex-cli should expect this thin workflow.sh-only shape, not steward's five-file layout. (Ref: steward#53.)

Prerequisites

Hard requirements: devex (the agentculture PR CLI; same tool as agex), gh (GitHub CLI — used only by the status / reply gating helpers, never to open a PR), jq, bash, python3 (stdlib only), curl (used by pr-status.sh).

Install devex once:

uv tool install devex   # or: pip install --user devex

Soft requirement: PyYAML is needed only for suffix mode of the sibling agent-config skill, where it parses Culture's server manifest. Every cicd script works without it; suffix mode prints a clear install hint when invoked without it.

Per-machine paths (sibling-project layout) live in .claude/skills.local.yaml; see the committed .example for the schema. devex pr delta reads the same file.

How to run

scripts/workflow.sh is the entry point. Subcommands:

Command	What it does
`workflow.sh lint`	`devex pr lint --exit-on-violation` — portability + alignment-trigger check.
`workflow.sh open [pr-flags]`	`devex pr open --delayed-read`. Creates the PR, then polls 180s for an initial briefing. `--title TITLE` required; body via `--body-file PATH` or stdin.
`workflow.sh read [PR] [--wait N]`	`devex pr read`. One-shot briefing (CI checks, SonarCloud gate + new issues, all comments, next-step footer). Pass `--wait N` to poll up to N seconds for required reviewers.
`workflow.sh reply <PR>`	`devex pr reply <PR>` — batch JSONL replies (stdin) + thread resolve. devex auto-signs from `culture.yaml`.
`workflow.sh delta`	`devex pr delta` — sibling alignment dump.
`workflow.sh status <PR>`	Steward extension. `pr-status.sh` — Sonar gate, OPEN issues, hotspots, unresolved-thread breakdown, deploy preview URL. Authoritative gate for `await`.
`workflow.sh await <PR>`	Steward extension. `devex pr read --wait` then `status`. Exits non-zero on Sonar ERROR or unresolved threads. Tunables: `STEWARD_PR_AWAIT_WAIT` (default 1800s passed to `--wait`), `STEWARD_PR_AWAIT_SECONDS` (legacy fixed pre-sleep, deprecated).
`workflow.sh help`	Print the list.

You can also call devex pr <verb> directly — workflow.sh is a typing-saver around the same verbs. The steward status and await extensions only have shell entry points.

The vendored single-comment helper pr-reply.sh (plus its _resolve-nick.sh dependency) is still shipped — pinned by tests/test_pr_reply_signature.py and tests/test_resolve_nick.py, and useful when a one-off reply doesn't merit batch JSONL. It is not called by workflow.sh anymore. The vendored portability-lint.sh is also still shipped — steward doctor's portability check runs it directly against target repos. Both are scheduled for follow-up migration to the CLI.

Long waits (background polling)

devex pr read --wait N polls in-session for up to N seconds. The Anthropic prompt cache has a 5-minute TTL; sleeping past it burns context every cache miss. Two ways to drive the wait:

Synchronous — workflow.sh await <PR> after workflow.sh open. Fine when readiness is expected within ~5 minutes.
Asynchronous — for longer waits, run devex pr read --wait NNN inside a background subagent (Agent tool, run_in_background: true) so the main session only pays the cache cost when readiness fires. The subagent's only job is to invoke devex pr read --wait and echo its headline back. The parent triages with workflow.sh await when the notification arrives. The user can interrupt with TaskStop.

This pattern was originally borrowed from sibling repo agentculture/cfafi's poll skill. The async guidance is also filed upstream (agex-cli#41).

Conventions

devex pr emits a "Next step:" footer at the end of every command that names the right next verb (the same chain devex learn cicd documents) — follow that rather than memorizing an order. workflow.sh help mirrors the verb table when you need the steward-flavored extensions (status, await) on top.

Branch naming: fix/<desc>, feat/<desc>, docs/<desc>, skill/<name>. PR / comment signature: - <nick> (Claude), where <nick> is resolved by devex from the agent's own culture.yaml (first agent's suffix), falling back to the git-repo basename. devex auto-appends the signature on pr open and pr reply only when the body isn't already signed.

Finishing a branch

When implementation on a branch is complete and tests pass, go straight to workflow.sh open (push the branch + open the PR). Do not stop to present a merge / PR / keep / discard menu and wait for a choice — in AgentCulture the standing default is always "push and create a Pull Request." This overrides the Superpowers finishing-a-development-branch skill's interactive menu, which otherwise pauses the flow at the moment work is ready to integrate. Merge-locally, keep-as-is, and discard happen only when the user explicitly asks for one of them.

Triage rules

For every comment, decide FIX or PUSHBACK with reasoning.

Default to FIX for: portability complaints (always valid for Steward — recurring bug class), test or doc requests, style nits aligned with workspace conventions.

Default to PUSHBACK for: architecture opinions that conflict with workspace CLAUDE.md or the all-backends rule; greenfield false-positives (e.g. "add tests" before there's any source — defer to a later PR, don't refuse).

Alignment-delta rule

If the PR touches CLAUDE.md, culture.yaml, or anything under .claude/skills/, run workflow.sh delta before declaring FIX or PUSHBACK on each comment. Note any sibling that needs a follow-up PR and mention it in your reply.

Greenfield-aware steps

The lint and the workflow script are always-on. Stack-specific steps are conditional and currently no-op (greenfield repo):

[ -d tests ] && [ -f pyproject.toml ] && uv run pytest tests/ -x -q
[ -f pyproject.toml ] && bump_version_per_project_convention   # see project README
[ -f .markdownlint-cli2.yaml ] && markdownlint-cli2 "$(git diff --name-only --cached '*.md')"

Revisit each line as the corresponding stack element actually lands. A pr lint --extra=tests,version,markdown ask is filed upstream (agex-cli#41).

Reply etiquette

Every comment must get a reply — no silent fixes. devex pr reply includes thread-resolve by default. Reference the review-comment IDs in the fix-up commit message.

The status extension queries SonarCloud directly (it predates the upstream Sonar integration in devex pr read). Both surfaces are trustworthy — devex pr read for display in the briefing, status for the gate. Steward isn't yet a registered mesh agent, so the post-merge IRC ping that Culture's pr-review includes is still skipped — that returns when Steward joins the mesh.