skills
个人 agent skill 库 — skills/ 为唯一真相源,agent 目录经 symlink 消费
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Check it deeper on your machine
It has changed since its scan on 11 Nov 2026.
As a Claude Code plugin, install it and then ask Claude Code:
/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/Dimon94/skills with scanskillsafety
Or from a terminal, clone it and run it:
git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/Dimon94/skills'
It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.
Badge for your README
Markdown
[](/listings/dimon94--skills)HTML
<a href="/listings/dimon94--skills"><img src="/api/listings/dimon94--skills/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
git clone https://github.com/Dimon94/skills.git \ "$HOME/.local/share/dimon-agent-workflow/skills" cd "$HOME/.local/share/dimon-agent-workflow/skills" ./scripts/install-development-workflow.sh ./scripts/install-development-workflow.sh --check
The script clones or fast-forwards standalone dependency repositories, then links repository-owned Skills, curated snapshots, and standalone Skills into `~/.agents/skills` and `~/.claude/skills`. It refuses to replace a real directory with a symlink. A dependency checkout with local changes is preserved and not updated.
To let an AI Agent inspect the environment, install runtime dependencies, and verify the result, paste the prompt from the [one-paste Agent bootstrap guide](docs/agent-bootstrap.md).
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 113 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.
- # Agent Development Workflow
- [中文](README.zh-CN.md)
- This repository combines requirement shaping, map planning, spec and ticket creation, automated dispatch, and local Git closeout into one resumable development workflow. `skills/` is the canonical runtime source. Third-party suites contribute only curated upstream snapshots; standalone single-Skill projects stay in upstream clones and are consumed through symlinks.
- [](docs/assets/development-workflow.en.svg)
- [View full-size SVG](docs/assets/development-workflow.en.svg) · [Download PNG](docs/assets/development-workflow.en.png)
- Follow the full chain from intake through large / regular / bug routing, MMF planning, ticket publication, implementation, automated dispatch and integration, current-head review, deployment, business acceptance, and closeout. Local Git closeout and remote delivery follow their own authority. See [Task Routing](#task-routing) for individual Skills.
- Run `python3 docs/assets/render-development-workflow.py` to rebuild the SVGs from the layout sources in `docs/assets/workflow-source/`, then validate and export PNGs with [fireworks-tech-graph](https://github.com/yizhiyanhua-ai/fireworks-tech-graph).
- ## Task Routing
- | Task type | Workflow |
- | --- | --- |
- | Large requirement | `to-epic` resolves planning questions, cuts the Epic into MMFs, and publishes the Epic Issue when authorized → `to-spec` writes one Spec per MMF → `to-tickets` cuts the MMF into a contract ticket plus concurrent behavior tickets that each own unique T/R items, sizes them with `ticket-sizing`, checks coverage and the dependency graph by script, and publishes after the user decides → implement; each MMF ships as one MR, and a one-ticket MMF is a one-row slice table |
- | Regular requirement | the model drafts a minimal plan → `grill-with-docs` stress-tests it → `implement` |
- | Small change | `implement` directly; `tdd` decides per change whether a failing check comes first |
- | Refactor or optimization | Analyze with either `improve-codebase-architecture` or `productionize-app-with-services` → classify the resulting work by size → follow the large- or small-requirement workflow above |
- | Bug | `git-worktree` → `diagnosing-bugs` → `git-commit` → `git-rebase-main` |
- | Understand existing code | `investigate` runs a read-only investigation: it routes mechanism questions to `how`, adds `why` for motivation, and answers decisions with a recommendation and tradeoffs |
- | Create or update a personal working-style Skill | run `/automate-me`: gather cross-session evidence, calibrate preferences, then write an independent `<handle>-mode` |
- | Assess a change's blast radius | `blast-radius` proves what a change breaks beyond the diff by running real code; foggy, wide impact goes to `wayfinder` |
- | No scripted way to prove a project works | Run `create-verification-skill` once per project to generate a local `verify-<app>` skill; `maintain-verification-skill` keeps its feature map honest |
- Every row that writes files starts in its own tree from `git-worktree`, even a one-line doc fix; the main checkout only integrates. `git-worktree` supports any number of task trees, symlinks `node_modules`/`.env`/`.venv` from main instead of reinstalling, and reclaims merged, clean, idle trees before creating a new one.
- Refactor and optimization analysis is also a pre-routing step, not an independent delivery path. Pushes, pull requests, and remote merges always require separate authority.
- When a pull request reaches review, `gh-merge-pr` freezes one Review Packet, runs `code-review` as the main Standards and Spec review, and adds specialist reviews only when the request, changed paths, or main findings provide evidence. The current-head verdict must pass before a separately authorized merge and parity readback.
- ## Install the Skill Bundle Once
- ```bash
- git clone https://github.com/Dimon94/skills.git \
- "$HOME/.local/share/dimon-agent-workflow/skills"
- cd "$HOME/.local/share/dimon-agent-workflow/skills"
- ./scripts/install-development-workflow.sh
- ./scripts/install-development-workflow.sh --check
- ```
- The script clones or fast-forwards standalone dependency repositories, then links repository-owned Skills, curated snapshots, and standalone Skills into `~/.agents/skills` and `~/.claude/skills`. It refuses to replace a real directory with a symlink. A dependency checkout with local changes is preserved and not updated.
- To let an AI Agent inspect the environment, install runtime dependencies, and verify the result, paste the prompt from the [one-paste Agent bootstrap guide](docs/agent-bootstrap.md).
- To deliver work with several coordinated agents after installing, follow the [multi-agent onboarding guide](docs/onboarding.md) (Chinese) from ticket slicing to merging back into main.
- ## Dependency Repositories
- | Repository | Workflow responsibility | Standalone install |
- | --- | --- | --- |
- | [Dimon94/skills](https://github.com/Dimon94/skills) | `implement`, `tdd`, `code-review`, `to-spec`, and local delivery Skills such as `git-commit` and `git-rebase-main` | [`scripts/link-skills.sh`](scripts/link-skills.sh) |
- | [mattpocock/skills](https://github.com/mattpocock/skills) | Research, grilling, architecture analysis, Wayfinder, and tickets | [Installation](https://github.com/mattpocock/skills#installation-30-second-setup) |
- | [swyxio/skills](https://github.com/swyxio/skills) | Upstream source for the curated `productionize-app-with-services` snapshot | Included by this repository's installer |
- | [cursor/plugins](https://github.com/cursor/plugins) | Upstream source for the `thermo-nuclear-code-quality-review` snapshot and the adapted `how`, `why`, `blast-radius`, `investigate`, `sharpen`, `correct`, and verification-skill pair | Included by this repository's installer |
- | [humanlayer/skills](https://github.com/humanlayer/skills) | Upstream source for the curated `show-me` snapshot | Included by this repository's installer |
- | [Kappaemme-git/codex-complexity-optimizer](https://github.com/Kappaemme-git/codex-complexity-optimizer) | Standalone `complexity-optimizer` Skill | Linked directly by this repository's installer |
- | [Dimon94/delivery-pipeline](https://github.com/Dimon94/delivery-pipeline) | Automated dispatch, integration, testing, and review from a Map or Spec | [Install](https://github.com/Dimon94/delivery-pipeline#install) |
- | [Dimon94/brainstorming-only](https://github.com/Dimon94/brainstorming-only) | `brainstorming-only` and `office-hours-only` | [Quick Install](https://github.com/Dimon94/brainstorming-only#quick-install) |
- | [fireworks-tech-graph](https://github.com/yizhiyanhua-ai/fireworks-tech-graph) | Generate and validate architecture, flow, and technical diagrams | [Installation](https://github.com/yizhiyanhua-ai/fireworks-tech-graph#installation) |
- | [Second Opinion](https://github.com/Dimon94/codex-with-chatgpt) | Optional high-level planning and independent review | [One-paste install](https://github.com/Dimon94/codex-with-chatgpt#one-paste-install--一段话安装) |
- | [Herdr](https://github.com/thinkerisme/herdr) | Terminal multi-agent runtime used by `delivery-pipeline` | [Install](https://herdr.dev/install) |
- `delivery-pipeline` also needs at least one installed worker CLI: pi, Codex CLI, or Claude CLI. Codex App uses native Tasks and Worktrees and does not require Herdr.
- ## Repository Boundary
- - `AGENTS.md`: repository working contract.
- - `GLOSSARY.md`: canonical vocabulary.
- - `docs/architecture/current-system-map.md`: modules, dependencies, and verification entry points.
- - `skills/`: the canonical source for repository-owned Skills and curated upstream snapshots.
- - `scripts/link-skills.sh`: links repository-owned Skills and upstream snapshots from `skills/`.
- - `scripts/install-development-workflow.sh`: installs the complete development-workflow Skill Bundle.
- - `skills/sync-upstream-skills/references/sources.json`: source, path, commit, and content hash for every upstream snapshot, plus `knownBrokenLinks` for links inside a snapshot that this repo does not import.
- - `skills/sync-upstream-skills/references/adapted-sources.md`: upstream source and base commit for repo-owned Skills adapted from upstream.
- Run `$sync-upstream-skills` to check curated snapshots; it writes only after an explicit update request. Standalone single-Skill projects still update through their upstream clone, and symlinks consume the result immediately.
- ## Principle Leaves
- `skills/principle-*` are pstack principle snapshots, locked by the `cursor-pstack` source in `sources.json`. Host Skills and docs point to a leaf at the decision it governs and never copy its text. See `docs/adr/0009-principle-leaves-and-index.md`. `dimon-mode` is the explicit-only working-style entry that lists every leaf (`docs/adr/0010-dimon-mode-entry.md`).
- | Leaf | Host |
- | --- | --- |
- | `principle-model-the-domain` | `to-spec` (Implementation Decisions) |
- | `principle-redesign-from-first-principles` | `to-spec` (Implementation Decisions) |
- | `principle-fix-root-causes` | `docs/agents/debugging.md` §4 |
- | `principle-separate-before-serializing-shared-state` | `to-tickets` (step 4, dependency edges) |
- | `principle-migrate-callers-then-delete-legacy-apis` | `to-tickets` (step 3, structure tickets; exception mode: check 6 in `references/release-per-ticket.md`) |
- | `principle-laziness-protocol` | `implement` (step 2, subtract first) |
- | `principle-subtract-before-you-add` | `implement` (step 2, subtract first) |
- | `principle-minimize-reader-load` | `implement` (step 4, ablation before reconciliation) |
- | `principle-test-behavior-not-implementation` | `tdd` (step 3, assertions survive internal changes) |
- | `principle-prove-it-works` | `implement` (step 3 exit, results read from the real artifact) |
- | `principle-build-the-lever` | `implement` (step 2, non-trivial slice gets a rerunnable script) |
- | `principle-encode-lessons-in-structure` | `retro` (step 1 recurrence, step 4 structural pass) |
- | `principle-never-block-on-the-human` | `dimon-mode` (自主权) |
- | `principle-boundary-discipline` | `AGENTS.md` `<principle_index>` (replaces the boundary line in `<constraints>`) |
- | `principle-sequence-verifiable-units` | `AGENTS.md` `<principle_index>` (replaces the minimal-slice line in `<principles>`) |
- | `principle-type-system-discipline` | `dimon-mode` (原则·架构) |
- | `principle-make-operations-idempotent` | `dimon-mode` (原则·架构) |
- | `principle-outcome-oriented-execution` | `dimon-mode` (原则·核心) |
- | `principle-explain-the-number` | `dimon-mode` (原则·验证) |
- | `principle-attack-the-premise` | `dimon-mode` `playbooks/bug-fix.md` (step 5, two failed fixes on one premise) |
- | `principle-foundational-thinking` | `dimon-mode` (原则·核心) |
- | `principle-exhaust-the-design-space` | `AGENTS.md` `<principle_index>` (index line carries the rule; upstream 08bd59b cut the body) |
- | `principle-experience-first` | `dimon-mode` (原则·核心) |
- | `principle-guard-the-context-window` | `dimon-mode` (子代理; `session-pickup` step 2) |
- Adding a leaf changes four places: the `cursor-pstack` source, the principle section of `dimon-mode`, the `<principle_index>` block in `AGENTS.md`, and this table. Import a leaf only when a host points to it.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.