Agent Projects and Skills for Software Development Workflows

Compare planning, agent orchestration, repository context, tool connections, and handoff resources for organizing software work.

Connect Planning, Execution, and Review

Development workflows organize the work around a code change: clarifying decisions, carrying context, assigning roles, connecting tools, and reviewing results. Harnesses and workbenches expose different execution mechanisms; Skills capture reusable procedures. Compare the stage you want to improve before selecting a runtime or adding another instruction to an existing agent.

Development Workflows Skills

Development Workflows Projects

Skills supply instructions; projects supply software or supporting resources. Check each source for setup, runtime, access, and license requirements.

Choose by the Job

The Plan Is Still Ambiguous

Use Grill Me or Brainstorming to clarify goals and decisions. The merged Grill Me guide also explains the document-aware mode for domain terminology and architecture records.

Components Are Growing Hard to Change

Composition guidance helps separate responsibility and avoid multiplying unclear prop combinations.

The App Needs a Real Check

Webapp Testing focuses on running and inspecting the application, not only reviewing source code.

Explore Related Task Categories

Find instructions and software for the next stage of your work, from code changes and interface design to office tasks and media production.

Frequently Asked Questions

How is this category different from hands-on Coding Skills?
Development covers decisions around the code change: understanding a system, clarifying a plan, designing components, and checking a running application. Coding focuses more narrowly on implementing, testing, or debugging the change itself.
What should I resolve before asking an agent to implement a plan?
Define the user outcome, constraints, affected interfaces, and how success will be checked. If a consequential decision is still open, surface it explicitly so the implementation does not silently turn an assumption into a requirement.
When is reverse engineering the right approach?
Use it when understanding a shipped application or binary is necessary and authorized, especially when source code is unavailable. Keep the investigation focused on a specific question and separate observed evidence from hypotheses about internal design.
When should I study a harness rather than add a Skill?
Study a harness when the question concerns tool execution, context handling, approvals, or agent coordination. Add a Skill when the host already supplies the needed capabilities but the task needs a clearer procedure. Neither choice implies that a project is already integrated with your workspace.
What does a real application check add to source review?
Running the application can reveal broken navigation, missing content, unexpected loading states, and interaction failures that source inspection misses. Confirm the expected user outcome, not just a successful response code or the absence of a compiler error.
How can I keep an agent-led development task from expanding indefinitely?
Write a concrete completion condition and separate necessary changes from follow-up opportunities. Review the changed surfaces against that condition, then stop when the agreed behavior and checks are satisfied.
What belongs in a useful handoff?
Record the goal, relevant files, decisions already made, checks completed, and the next unresolved action. A new agent should be able to distinguish verified progress from assumptions without reconstructing the entire conversation.
When should a development workflow become a reusable Skill?
Consider it when the same task has a stable input, repeatable method, and a result you can evaluate. Start with a small instruction and representative examples, then revise it using observed outcomes instead of packaging every one-off task.

Keep Development Decisions Visible

Work with your files, implementation notes, and verification in one place.