project-setup
Use when starting work on a new or existing project that lacks a `.devin/` configuration, or when the user wants to add skills, hooks, rules, and local tools to a project in a systematic, validated way.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- project-setup
- version
- 1.0.0
- triggers
- [ "user", "model" ]
- description
- Use when starting work on a new or existing project that lacks a `.devin/` configuration, or when the user wants to add skills, hooks, rules, and local tools to a project in a systematic, validated way.
02 资源文件 1
- 参考文档templates/agents.md1.4 KB
03 SKILL.md 正文
Project Setup
Set up a Devin CLI project workspace inside .devin/ only. This is a systematic, validated workflow that adapts the continuous-improvement 10-step loop to project onboarding.
Principle
Every project that the agent touches deserves a predictable, auditable .devin/ setup. Nothing agent-facing is written outside .devin/. The skill does not write business code or install project dependencies unless they are required for the agent to function.
When to use
- The repo has no
.devin/directory. .devin/exists but is incomplete (no rules, no hooks, no skills, no memory, no agent config).- The user asks to "set up the project for the agent", "add skills/hooks", or "prepare this repo for Devin CLI".
When NOT to use
- The project already has a complete
.devin/and the user wants a specific edit - use/self-extendinstead. - The task is to write application code - use
/implementor/tdd. - The user only wants a global config change - use
~/.config/devin/paths.
FASE 0 - Deep Research (before any change)
Run these checks in order and record the evidence. Do not proceed until each step has a concrete output.
-
Inspect the repo.
git remote -vand.git/config- GitHub, GitLab, or none?ls -la(orGet-ChildItem) at the repo root - what files and directories already exist?- Look for
package.json,pyproject.toml,requirements.txt,Cargo.toml,pnpm-workspace.yaml,go.mod,pom.xml, or similar stack signals. - Look for
src/,packages/*,apps/*,tests/to detect monorepo.
-
Inspect existing
.devin/(if any).ls -la .devin/- what is already configured?- Read
.devin/config.json,.devin/hooks.v1.json,.devin/mcp_config.jsonif they exist. - List
.devin/skills/*/and.devin/rules/*.md.
-
Confirm Devin CLI conventions.
- Rules:
.devin/global_rules.mdand.devin/rules/*.md(sources: Devin CLI docs,self-extendskill). - Skills:
.devin/skills/<name>/SKILL.md. - Hooks:
.devin/hooks.v1.json. - MCP:
.devin/mcp_config.jsonand.devin/mcp_config.local.json(gitignored). - Memory:
.devin/memory/(see/project-memory).
- Rules:
-
List the bundle resources available.
- Read the global bundle
config.json,manifest.json, andAGENTS.mdto know which skills, scripts, and hooks can be copied into the project.
- Read the global bundle
LOOP DE SETUP (10 passos)
Passo 1 - OBSERVAR
Record the current state in a ledger at .devin/ledgers/project-setup.md (create the ledgers/ directory if it does not exist):
# Project setup: <repo-name>
## Observed
- Stack: <detected stack>
- Monorepo: <yes/no>
- Remote: <github/gitlab/none>
- Existing `.devin/`: <list files>
## Missing
- <rule file>
- <hook file>
- <skill file>
- <memory dir>
Passo 2 - CRITICAR
Compare the observed state against the rule: all agent files must live in .devin/. Identify every missing or mis-placed item. For each missing item, note which bundle resource should provide it.
Passo 3 - GERAR ALTERNATIVAS
For each missing component, generate at least two alternatives and record trade-offs:
| Component | Alt 1 | Alt 2 | Alt 3 | Best fit |
|---|---|---|---|---|
| Issue tracker | GitHub (gh) | Local markdown | GitLab | depends on git remote |
| Rules | .devin/global_rules.md |
.devin/rules/*.md |
both | both |
| Memory | .devin/memory/ |
none | - | always enable |
| Hooks | copy bundle hooks | create minimal set | none | copy from bundle |
| MCP | .devin/mcp_config.json template |
none | - | ask user |
Passo 4 - REVISAR
Apply the best-fit alternative for each component. This is the deterministic setup order:
- Run the engineering-skills setup to configure issue tracker, triage labels, and domain docs (all inside
.devin/). - Create
.devin/global_rules.mdwith a short## Agent skillsblock and the repo-specific rules from the engineering-skills setup. 2.5. Create.devin/rules/agents.mdfrom the project template - copyskills/project-setup/templates/agents.mdinto.devin/rules/agents.mdand replace the stack-specific placeholder with the project's concrete rules (e.g., TypeScript withoutany, linter commands, naming conventions). This is the project's per-stack agent rules file; keep it short and action-oriented. - Create
.devin/rules/*.mdfor optional, trigger-scoped rules. - Create
.devin/hooks.v1.jsonwith the essential hooks from the bundle:behavioral-nudge.pyonUserPromptSubmitcheck-ai-signature.pyonPreToolUseforwrite/editcheck-push-green.pyonPreToolUseforexecdestructive-gate.pyonPreToolUseforexeccontext-budget.pyonSessionStartmemory-retrieval.pyonUserPromptSubmitmemory-post-edit.pyonPostToolUseforwrite/editmemory-post-exec.pyonPostToolUseforexecmemory-stop.pyonStop
- Create
.devin/mcp_config.jsonas an empty scaffold if the project has no MCP servers; otherwise ask the user which servers to add. - Create
.devin/skills/project-memory/if it does not exist, copyingnote.md,capture-memory.py,query-memory.py, andaudit-memory.pyfrom the bundle. Use/project-memoryto walk the user through the first capture. - Create a local copy of the engineering-skills setup if the user wants the engineering flow available locally.
- Create
.devin/memory/directory and seedMOC.md.
Passo 4.5 - POPULAR A BASE DE CONHECIMENTO
The .devin/ structure is not a solid base until it has content. Run the following setup skills in order, recording everything in .devin/ledgers/project-setup.md:
- Run
grillingto capture the project's big-picture goals, constraints, and stakeholders. Save the approved design (if one emerges) to.devin/specs/. - Run
domain-modelingto build.devin/CONTEXT.mdand.devin/adr/. Resolve ambiguous or overloaded terms before they leak into code. - Run
project-memoryto capture a first memory entry - stack, conventions, and any decisions the user already shared. - Run
triageonce on the existing issue tracker to show the user the triage flow and confirm label mapping. - Run
planning-pipelineon the most urgent item to demonstrate the spec -> tickets flow.
Stop after each step if the user has no content to populate yet. The goal is to leave the project with at least a populated CONTEXT.md, one memory entry, and a triaged issue, not to force a full spec.
Passo 5 - VALIDAR
Run the project-level validation suite:
python audit.pyin the project root if the project has one.- If no project audit exists, run a minimal structural check:
ls .devin/global_rules.md,.devin/hooks.v1.json,.devin/mcp_config.json,.devin/skills/,.devin/memory/python -m json.tool .devin/hooks.v1.jsonpython -m json.tool .devin/mcp_config.json
Passo 6 - FUTURE PACE
Project the setup against 3 hypothetical future sessions:
- New feature request - does the agent know where to write tickets, capture memory, and read domain docs?
- Code review request - does the agent find the standards sources and the
AGENTS.mdrule block? - Multi-session task - does the agent have memory retrieval hooks and a local issue tracker to record progress?
At least two must be clearly helped; otherwise the setup is incomplete.
Passo 7 - ECOLOGICAL CHECK
Check for side effects:
- Did the setup create any file outside
.devin/? If yes, delete or move it. - Did any hook conflict with an existing project hook?
- Is the
context-budget.pyhook adding too much token cost? - Are there duplicate skills in
.devin/skills/and%APPDATA%/devin/skills/?
Passo 8 - SIMULAR
Run install.ps1 -Force (or copy the .devin/ files) and restart a fresh Devin CLI session against the project. Verify:
skill list --path .devin/skillsshows the local skills.python audit.py(or the structural check) passes.- A test
writeto a project file triggerscheck-ai-signature.py(if configured).
Passo 9 - CLASSIFICAR
Classify the setup:
| Classe | Critério | Ação |
|---|---|---|
| MELHOROU | All 3 future scenarios helped, no side effects, validation passed | Mark ledger as done |
| NEUTRO | Validation passed, no side effects, but <2 scenarios helped | Revisit missing component |
| PIOROU | Validation failed or side effects found | Revert the offending change and return to Passo 3 |
| INCONCLUSIVO | Could not simulate (no fresh session) | Do not declare done; ask user to test |
Passo 10 - REPETIR OU CONVERGIR
- If the user asks for more tools, skills, or hooks, return to Passo 1 and update the ledger.
- If nothing more is requested and the setup is classified MELHOROU or NEUTRO, the setup is complete.
Output format
When finishing, report:
SETUP: <repo-name>
OBSERVED: <stack> | <monorepo> | <remote>
FASE0: <sources checked>
COMPONENTS: <list of .devin/ files created or updated>
VALIDATION: <pass|fail>
FUTURE_PACE: <n/3 scenarios helped>
ECOLOGICAL: <side effects or none>
CLASSIFICATION: <MELHOROU|NEUTRO|PIOROU|INCONCLUSIVO>
PENDING: <what the user still needs to decide or provide>
Red flags - STOP and ask
- The user asks to write files outside
.devin/. - A requested hook conflicts with the project's CI or security policy.
- The user wants to skip the engineering-skills setup but still use
triage,planning-pipeline, orwayfinder. - A proposed MCP server requires a secret that the user has not provided.
Common mistakes
- Copying every global skill into the project. Only copy skills that benefit from project-specific tuning; the global bundle is still available.
- Creating
AGENTS.mdat the project root. Always prefer.devin/global_rules.md. - Skipping the ledger. Without a ledger, the agent cannot prove completion.
- Skipping simulation. A setup that is not loaded by a fresh session is not done.
Cross-references
- The engineering-skills setup - issue tracker, triage labels, domain docs.
/grilling- capture the project's goals and produce a design doc./ai-coding-dictionary- align any AI-coding jargon before it enters the project glossary./domain-modeling- build.devin/CONTEXT.mdand.devin/adr/./project-memory- capture and retrieve project knowledge./triage- first triage run on the issue tracker./planning-pipeline- demonstrate spec → tickets flow./self-extend- add individual rules, skills, hooks, or MCP servers./continuous-improvement- the 10-step loop this skill adapts.
04 同仓库更多 Skills
montador-de-ideias
Use when assembling, expanding, or elevating a raw project idea or initial prompt into an ultra-detailed, first-principles dossier and compiled master prompt with journalistic OSINT market research, Tech Lead and Staff Architect justifications, atomic and modular hardening/optimization, and high-impact contact surfaces.
git-github-adaptive-engineering
Use when any task touches git history, GitHub issues/PRs/releases/CI, repo governance, or branch and merge decisions. Runs the Fase Zero situational-awareness battery, picks one of five delivery flows via the adaptive decision engine, and enforces commit/PR information-density standards. Orchestrates the gh, git-workflows, github-actions, pr, and debug/finish skills.
adaptive-agenda-calendar-routine-engine
Use when planning or executing work across hours, days, weeks, or sprints. Turns the calendar into a strict time-blocked agenda (HH:MM-HH:MM blocks carrying a routine profile, expected inputs, and a handoff output artifact), runs the 5-stage cognitive flow (Radar and Hunt -> Triage plus Steve Jobs conception -> workbench manifest -> >=32-cycle loop -> Close plus Open-Minded Scrum), and unfolds macro calendar milestones into micro agenda blocks through extensions/gcal/wrapper.py.
demo
Use when demonstrating a skill in a fixture project.
duplicate-skill
Use when testing duplicate skill name detection.
duplicate-skill
Use when testing duplicate skill name detection.