- Home
- Skills
- Prompt & Skill Engineering
- Builder Kit
Works with the AI Agents you already use
Builder Kit
Plan, build, verify, and review software changes with repository-aware AI workflows, approval gates, and evidence-backed checks. Free and open source.
Free
Builder Kit
Example session with this skill installed
Use Builder Kit to plan a profile-update API endpoint for our Python backend.
Users should be able to update their display name and bio, but not their email, password, role, or account status.
Give me a proposed API contract, security requirements, implementation plan, and acceptance criteria. Mark anything that depends on repository inspection as unverified. Do not change files until I approve the plan.
- Read your context and instructions
- Compiled the builder kit
Summary
Classification: planned work. This feature introduces an API endpoint and authorization requirements, so it needs an approved plan before implementation.
Selected lenses: backend engineering, security and privacy, and quality engineering.
No repository has been inspected. The proposal below must be checked against the application's existing routes, authentication, validation, and persistence conventions. No files have been changed.
Proposed API contract
Endpoint: PATCH /users/me
This path is provisional; reuse the repository's established API prefix and naming conventions.
Example request
```json
{
"display_name": "Alex Morgan",
```
"bio": "Developer building useful tools."
```
}
| Field | Proposed behavior |
|---|---|
| `display_name` | Optional in a partial update; when supplied, must be a non-empty string |
| `bio` | Optional in a partial update; an empty string clears it |
| Omitted fields | Remain unchanged |
| Unknown or protected fields | Reject the request rather than silently ignoring them |
Maximum lengths, whitespace handling, null handling, and empty-request behavior must be confirmed against existing product rules and model constraints.
**Proposed success response**
~~~json
```json
{
"display_name": "Alex Morgan",
"bio": "Developer building useful tools."
}
~~~
Use the application's existing response envelope and status conventions if they differ. Return no sensitive account fields.
Security requirements
- Resolve the target account from the authenticated principal. Do not accept a target user ID in the request body.
- Explicitly allow only
display_nameandbio; do not assign arbitrary request properties to the user model. - Reject attempts to change email, password, role, or account status.
- Apply existing authentication, CSRF protection where relevant, and request-size controls.
- Preserve existing output-escaping rules for profile text; accepting text does not make it safe to render as HTML.
- Avoid logging credentials, session tokens, or unnecessary profile content.
These are implementation requirements, not verified properties of the current application.
Implementation plan
- Inspect repository instructions, related endpoints, user models, authentication helpers, and tests.
- Confirm field constraints, error formats, and update/concurrency conventions; revise this proposal where necessary.
- Obtain approval for the finalized contract and implementation plan.
- Add the request schema and endpoint using existing application patterns.
- Update only supplied, permitted fields through the established persistence layer.
- Add focused tests, run relevant checks, and report actual outcomes and limitations.
Acceptance criteria
| Scenario | Expected result |
|---|---|
| Authenticated user updates both permitted fields | Both values persist and appear in the response |
| User updates only their bio | Display name remains unchanged |
| User supplies an empty bio | Bio is cleared |
| Request is unauthenticated | Existing authentication error; no mutation |
Request includes role or another forbidden field | Validation error; no partial mutation |
| Request violates confirmed field constraints | Existing validation error format; no mutation |
| Update succeeds | Response contains only the approved profile fields |
Next steps
Provide access to the repository so its conventions and constraints can be checked. Any material differences will be reflected in the final plan before approval.
Status: proposal prepared; repository-dependent details unverified. No implementation or verification has been performed.
Connects securely to your tools. The creator never sees your data.
About this skill
Builder Kit gives your coding agent a structured workflow for changing software: understand the repository, classify the task, plan substantial changes, get approval, implement, and verify.
It brings together five workflows—Builder Kit Router, Develop from Spec, Start New App, Adversarial Review, and PR Review Loop—with shared engineering references, planning templates, and Git safety guidance.
Free and open source under the MIT license.
What it does
-
Matches process to risk. Separates localized tiny fixes from substantial work that needs specification, planning, and approval.
-
Loads relevant guidance. Uses architecture references and specialist lenses for data, APIs, security, reliability, accessibility, and frontend work.
-
Sets explicit safety boundaries. Instructs the agent to preserve unrelated changes, respect repository rules, and obtain authorization for publishing or destructive actions.
-
Challenges assumptions. Reviews specifications, plans, architecture, and code for material flaws, distinguishing evidence from uncertainty.
-
Reviews and fixes iteratively. When fixes are authorized, repeats review and verification within a bounded loop, reporting unresolved risks and blockers.
-
Guides new applications. Works through product discovery, design decisions, approval, and scaffolding without presenting a prototype as production-ready.
How it works
-
Ask explicitly. For example: “Use Builder Kit to plan this feature. Do not implement until I approve.”
-
Discover and classify. The agent reads repository instructions and relevant code, identifies constraints, and selects the appropriate workflow.
-
Plan and approve. Non-trivial implementation requires an approved plan. Large or high-risk work requires specification approval before planning.
-
Implement and verify. Authorized changes proceed in coherent slices with proportionate checks and a clear account of results and limitations.
Tiny fixes use a lighter path. Adversarial reviews remain read-only unless changes are separately authorized.
Languages and tools
The workflows are language-agnostic, with additional reference guidance for Java, Python, C++, Jenkins, and frontend development.
Repository work uses the host agent’s available filesystem, shell, and Git tools. GitHub-specific actions require suitable tools, authentication, and explicit authorization; this package does not supply those capabilities.
Why use it instead of a one-off prompt?
Builder Kit makes engineering expectations reusable: what to inspect, when to stop for approval, which risks to consider, and what evidence to gather before claiming completion.
Its value is a consistent process grounded in the repository—not a promise that an agent will never make mistakes.
Example requests
-
“Use Builder Kit to specify and plan a new API endpoint.”
-
“Use Builder Kit to investigate this localized bug and verify the fix.”
-
“Run a read-only adversarial review of this architecture.”
-
“Run the PR review loop on my current changes. Ask before editing.”
-
“Use Start New App to discover requirements and plan a new application.”
Requirements and limitations
-
The Agensi connector must make the bundled supporting documents accessible to the host agent. If a required resource cannot be retrieved, the affected workflow should stop and report it.
-
Implementation and verification require authorized editing and command-execution tools.
-
This listing delivers workflow instructions and reference materials, not a hosted coding service or native plugin installation.
-
Instructions cannot guarantee agent compliance, defect-free code, security, or production readiness.
-
End-to-end execution through Agensi has not yet been verified.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
Requires host agent filesystem and Git tool access. Workflow halts if required reference documents are missing. Maximum 10 iterations for the PR review loop.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
2 uses
Times used to date
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 27 days ago
- Free with an Agensi account
- Accounts
GitHub
OpenAI
Anthropic
Permissions
Allowed Hosts