feat(adapter/aider): Aider skill runtime
- Dominant language
- Python
- Stars
- 92
- Forks
- 92
- Avg merge
- 1d 12h
- Merged PRs (30d)
- 104
Description
**Goal:** Bring Aider to feature parity with Claude Code as a skill runtime. Grounds: [RFC-AI-0004 Principle 3](../tree/main/docs/rfcs/RFC-AI-0004.md) + the RFC's adoption checklist, which names Aider as a host with per-action confirmation primitives.
**Why Aider:** Mature, fully open-source, large active community, multi-model (works with Claude, GPT, Gemini, local LLMs, OpenRouter — the most vendor-agnostic agent runtime in the field). Aider is the natural fit for adopters who want to keep their LLM choice open without committing to one vendor's CLI.
**What "parity" means:**
- Skills under `.claude/skills//SKILL.md` are invokable as Aider workflows (`/run`, custom slash commands, scripted Aider sessions)
- The `tools/*` bridges are reachable via Aider's shell-command primitives
- Sandbox parity: Aider doesn't ship its own sandbox; the framework's `bubblewrap` baseline wraps Aider directly — document the recipe
- HITL primitives: Aider's "approve every edit" is the native fit; verify and document for the framework's "draft before send" rule
**Suggested approach:**
- Map `SKILL.md` → Aider's `CONVENTIONS.md` + `.aider.conf.yml` + per-session `/read`-loaded prompt files
- Document how to compose a skill-as-Aider-workflow: read the SKILL.md, follow the procedure, call the bridges
- Validate against the existing eval suite ([`tools/skill-evals/`](../tree/main/tools/skill-evals/)) — Aider's multi-model nature makes it a useful canary for "does this skill *actually* work cross-model?"
**Reference:**
- RFC-AI-0004: [`docs/rfcs/RFC-AI-0004.md`](../tree/main/docs/rfcs/RFC-AI-0004.md)
- Aider: https://aider.chat
- `CONVENTIONS.md` convention: https://aider.chat/docs/usage/conventions.html
- Existing skill shape: [`.claude/skills/`](../tree/main/.claude/skills/)
Contributor guide
Assessment
This issue has not been assessed yet.