Onboard legacy project
Skill barry166/agent-stack-skills/skills/onboard-legacy-project
Orchestrates the onboarding workflow for an existing or unfamiliar software project by coordinating project mapping, environment bootstrap, and legacy guardrails.From its SKILL.md
npx -y skills add barry166/agent-stack-skills --skill onboard-legacy-projectAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 28 days oldThe repository was created 28 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.3 KB, 868 tokens by cl100k_base, as published. Nobody here has run it
Onboard Legacy Project
Guide a developer through legacy project onboarding without duplicating the underlying skills.
Default sequence:
map-legacy-project: build static project understanding.bootstrap-project-environment: make the project reproducible and smoke-tested.build-legacy-guardrails: add the minimum pre-change safety net.
Hard Rules
- Do not merge or rewrite the three downstream skills.
- Do not copy downstream workflow details here; load and follow the selected skill when execution begins.
- Before execution, confirm the selected downstream skills are installed or otherwise available; if any required skill is missing, stop and report the missing skill name.
- Default to recommending the next stage first, then wait for the user to confirm before running it.
- Stop after each completed stage for RCD and user confirmation.
- Continue through all stages only when the user explicitly asks for one-shot/full/autonomous completion.
- In one-shot mode, still run RCD after every stage and adjust the next stage from the produced evidence.
- Prefer the normal order:
map-legacy-project->bootstrap-project-environment->build-legacy-guardrails.
Stage Recommendation
First inspect the request and any existing project assets. Recommend the smallest useful path:
- New to the project, wants to understand it: recommend
map-legacy-project. - Wants to run or hand off setup: recommend
map-legacy-projectthenbootstrap-project-environment. - Wants to refactor, migrate, or safely modify code: recommend all three stages.
- Docs already exist and the user only worries about drift: recommend the project-local
docs-auto-syncskill if present.
If the user has not asked for one-shot execution, respond with the recommended stage/path and ask for confirmation in plain language. Do not start the downstream skill until the user confirms.
Execution Flow
When the user confirms:
- Load and execute the selected downstream skill.
- Run the RCD check for the completed stage.
- Report the RCD result and the recommended next stage.
- Pause for user confirmation unless one-shot mode is active.
Use existing outputs as inputs for later stages. For example, bootstrap-project-environment should reuse dependency and architecture facts from map-legacy-project; build-legacy-guardrails should reuse environment docs and critical project maps where they exist.
RCD Check
Run this after every stage:
- Review: expected files exist, required sections are present, and the stage's own completion criteria are met.
- Consistency: outputs agree with source evidence and with previous-stage assets; note anything that may poison later stages.
- Decision: choose one of
continue,fix-current-stage,skip-next-stage, orask-user, with a short reason.
Minimum stage expectations:
map-legacy-project: architecture/API/data-model docs,AGENTS.md, localdocs-auto-sync, and summary decision are present or explicitly skipped.bootstrap-project-environment: environment checklist, install/startup/smoke logs, setup guide, and safe skipped checks are documented.build-legacy-guardrails: critical paths, test inventory/gaps/plan, P0 guardrail result, CI decision, and summary are documented.
If RCD finds a clear fix inside the just-completed stage, fix it before moving on. If the issue is a product or business decision, stop and ask the user.
One-Shot Mode
One-shot mode is active only when the user explicitly requests wording such as "一次性完成", "全自动跑完", "不要中途停", "one shot", or "full onboarding".
In this mode:
- Execute stages in the normal order.
- After each stage, perform RCD silently enough to keep momentum but record the decision in the final report.
- If a stage fails or produces weak evidence, adapt the next stage instead of pretending it succeeded.
- Stop only for destructive actions, missing required authority, repeated blockers, or decisions that cannot be safely defaulted.
Final Report
Keep the final report short:
- stages run
- RCD decisions
- produced or reused files
- blockers or human-confirmation items
- recommended next action
What ships with it: 1 file
333 B alongside SKILL.md
agents/
- openai.yaml333 B