Planning & Workflow

Generate three to five distinct candidate approaches with their weakest links visible

Frame an engineering problem before choosing a solution, with constraints and acceptance

Set up Haft and build its initial project profile and specification map

Work through Greptile review feedback on a PR in bounded review and fix cycles

Run the greploop review loop on large PRs by triggering the Greptile apps bot

Isolate each feature or fix in its own git worktree before editing any repo files

Archive a shipped spec and rewrite it as a durable rationale record

Map the known and unknown parts of an ambiguous task in four quadrants

Hand off a spec that is mid-implementation to another machine or agent

Implement long specs in committed passes with periodic cleanup checkpoints

Have Codex implement a spec while you orchestrate, integrate and review

Implement a spec using the goal feature of the agent harness

Run a retrospective on a coding session and capture what to change

Closeout pass that runs refactor, code review and docs before you commit

Break large features into verifiable slices with specs and staged plans

Rewrite a finished commit message so git log tells the story clearly

Review and land external pull requests with credit, gates and clear explanations

Open a seeded casebook for a new build cycle before the first iteration

Restart a project from v0 while keeping only the proven lessons and tests

Design systems with module boundaries, API contracts and architecture decision records

Resolve git merge conflicts without losing the intent of either branch

Run one task through a V-model loop with verification the builder cannot grade

Run a retrospective on checkpoints, evidence quality and follow-up action items

Drive a large multi-session goal through staged closed-loop verification phases

Compare frameworks and architectures to settle a technology choice

Connect Fabric workspaces to Git and manage commits, sync and conflicts

Write a standalone bug report another developer can reproduce and fix

Set up GitHub auth with HTTPS tokens, SSH keys, credential helpers or gh

Create, triage, label and close GitHub issues and link them to pull requests

Create branches, open pull requests, watch CI, auto-fix failures and merge

Write a markdown plan for a coding task before any code is changed

Turn a multi-step spec into a plan with exact file paths and code

File a GitHub bug report or feature request from feedback you collect

Apply TRUST 5 quality principles, SPEC-first DDD and agent delegation patterns

Work through architecture choices and root causes with first-principles framing

Write SPEC documents with EARS requirements and acceptance criteria

Apply branch strategies, conventional commits and PR templates for git work

Implement SPEC requirements with DDD or TDD methodology and tests

Sync docs and codemaps, then open a pull request for the finished work

Orchestrate SPEC workflows from plan to run to sync with acceptance criteria

Generate boilerplate files and scaffolding from reusable project templates

Set up git worktrees so several SPECs can be built in parallel

Run the quality gate before committing, delegating checks to one entry point

Compare, validate and publish workspace artifact candidates with the Workbench CLI

Interview the user about a plan until every branch of the design is settled

Set up MATLAB projects with matlab.toml for dependencies, paths and shortcuts

Export session MATLAB code into a clean, runnable .m script

Share MATLAB code via GitHub or File Exchange with Open in MATLAB Online links

Work through claimed Chorus tasks, report progress and run parallel workers