Skip to content

agentsope/SkillAlchemy / skills/agentsop-aider

agentsop-aider

Verified

SOP for terminal-based, git-native AI pair programming with Aider (git work-tree + tree-sitter repo-map + edit-format + human-in-loop REPL). Use when editing code in an existing git repo via an LLM, when you need to converge a change to 2-5 files, pick an edit format that fits the model, run architect+editor mode, or wire an auto-test loop.

references/R5-ecosystem-context.md

View on GitHub
# R5: Ecosystem Context — Aider vs. Cline / Cursor / Continue / OpenHands

## The landscape, framed by interaction model

```
                  HUMAN-IN-LOOP                           AUTONOMOUS
                       |                                       |
  +-------------------+-+----------+----------+----------+----+----+
  | Terminal REPL     | IDE inline | IDE chat | Closed   | Sandbox |
  | (you steer turns) | (live      | (separate| IDE      | (no UI) |
  |                   |  cursor)   |  panel)  |          |         |
  +-------------------+------------+----------+----------+---------+
  | AIDER             | CURSOR     | CONTINUE | CURSOR   | OPENHANDS|
  |                   | / CLINE    |          |          |          |
  +-------------------+------------+----------+----------+----------+
```

Aider stakes out **terminal + human-in-loop**. Every other tool trades one of those axes.

## Side-by-side

| Aspect | Aider | Cline | Cursor | Continue | OpenHands |
|---|---|---|---|---|---|
| Surface | Terminal REPL | VS Code extension | Closed-source IDE (VS Code fork) | VS Code + JetBrains extension | Web UI + sandboxed Docker |
| Open source | Yes (Apache 2.0) | Yes | No | Yes | Yes (MIT) |
| Editing primitive | Diff/udiff/whole text + git commit | Tool calls with human approval per step | Composer agent, multi-file edits inline | Edit/Chat/Agent/Autocomplete modes | Autonomous plan→edit→test→PR |
| Context strategy | tree-sitter repo-map + selective /add | Reads files on demand via tool calls | Full repo indexing | RAG over repo | Reads files agentically |
| Human approval | Per turn (you press enter / `/undo`) | Per tool call (approve each edit/cmd) | Per Composer apply | Per edit | None (autonomous) |
| Git integration | Auto-commit per edit (native) | Through terminal tools | Manual | Manual | Sandboxed, returns PR |
| Bring-your-own model | Yes (100+ via LiteLLM) | Yes | Limited | Yes | Yes |
| Best for | Refactors with clear scope, surgical edits, agent-friendly scripting | Step-by-step approval-gated workflows in VS Code | Visual flow, autocomplete-heavy editing | Multi-IDE teams | Ticket-to-PR on isolated tasks |

Sources: [frontman.sh/blog/best-open-source-ai-coding-tools-2026], [cline.bot/blog/top-9-cursor-alternatives-in-2025], [opensourcealternatives.to/blog/best-open-source-ai-coding-assistants-2026], [shakudo.io/blog/best-ai-coding-assistants].

## When to reach for Aider over the alternatives

### Reach for Aider when:
1. **You live in the terminal.** tmux + vim/emacs + Aider is the canonical setup. No IDE switch cost.
2. **You want git-clean history.** Aider's per-edit commit pattern produces an auditable trail another tool would require manual discipline to match.
3. **You're scripting an agent.** Aider can be driven with `--message`, `--yes`, `.aider.conf.yml`. The terminal interface is naturally compose-able.
4. **You care about edit-format quality.** Aider has the most published research on edit formats (diff, udiff, whole, patch, JSON-vs-text). If your model has a known weak edit format, Aider has a path.
5. **You want a deterministic context strategy.** The tree-sitter repo-map is inspectable (`/map`), reproducible, and free of vector-DB infrastructure.
6. **The task is well-scoped.** You know which 2–5 files matter.

### Reach for Cline when:
- You're already in VS Code and want **per-tool-call human approval**. Cline shows every file edit and every shell command for confirmation. Aider has `/undo` but Cline has prevention.
- You want a visible step list as the agent works.

### Reach for Cursor when:
- You want **inline diff overlays in the editor**, autocomplete-style ghost text, and visual file-context indicators. Cursor's UX is a strength Aider doesn't pretend to match.
- You can pay for a closed product and want a polished single-vendor experience.

### Reach for Continue when:
- You need cross-IDE consistency (VS Code + JetBrains).
- You want **autocomplete + chat + agent in one extension**, with team-shared config.

### Reach for OpenHands when:
- You want **autonomous ticket→PR**. The agent reads an issue, plans, edits in a sandbox, runs tests, opens a PR. Aider does not target this.
- You're benchmarking on SWE-Bench-style tasks where autonomy is the metric.

## What Aider uniquely teaches the agent-coder

Even if you choose Cline or OpenHands as the surface, Aider's published research is the reference text for several decisions:

1. **Edit format selection** — the 20%→61% udiff jump on GPT-4 Turbo refactor benchmark is the canonical proof that format choice rivals model choice. [aider.chat/2023/12/21/unified-diffs.html]
2. **Repo-map over RAG** — 70.3% correct-file selection from a tree-sitter map on SWE-Bench Lite — beats many embedding-based retrievers. [aider.chat/2024/05/22/swe-bench-lite.html]
3. **Architect+editor split** — the 79.7%→85% jump on Polyglot was a public proof that two cheap calls beat one expensive call. [aider.chat/2024/09/26/architect.html]
4. **JSON-vs-text** — every model tested did worse with JSON-wrapped code. Caution for anyone designing tool-call interfaces. [aider.chat/2024/08/14/code-in-json.html]
5. **Lazy-coding mitigation** — the unified-diff prompt + flexible patching reduced "lazy comments" 3× without changing the model. Engineering wins over model choice. [aider.chat/2023/12/21/unified-diffs.html]

## Combination patterns

- **Aider + Claude Code / OpenHands**: orchestrator agent uses Aider as its editing tool. Aider's terminal + git interface composes well as a subprocess.
- **Aider + Cursor**: edit visual UI stuff in Cursor; switch to Aider for surgical refactors and chain edits.
- **Aider + IDE diagnostics**: pipe `tsc --noEmit` or `pyright` output via `/run` so Aider sees type errors it can fix.

## References

- [aider.chat/docs/leaderboards/]
- [aider.chat/2024/09/26/architect.html]
- [aider.chat/2024/05/22/swe-bench-lite.html]
- [aider.chat/2023/12/21/unified-diffs.html]
- [aider.chat/2024/08/14/code-in-json.html]
- [frontman.sh/blog/best-open-source-ai-coding-tools-2026]
- [cline.bot/blog/top-9-cursor-alternatives-in-2025]
- [shakudo.io/blog/best-ai-coding-assistants]
- [opensourcealternatives.to/blog/best-open-source-ai-coding-assistants]
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 322kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 322kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 322kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall