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.

intermediate/operation_candidates.json

View on GitHub
{
  "project": "Aider",
  "version": "1.0.0",
  "extracted_at": "2026-05-19",
  "dimensions": [
    {
      "id": "R1",
      "name": "Architecture & Mental Model",
      "operations": [
        {
          "operation": "treat repo as source of truth, chat as ephemeral",
          "trigger": "session start",
          "decision_rule": "files + git history persist; chat clears with /clear or /reset",
          "evidence_cite": "aider.chat/docs/git.html"
        },
        {
          "operation": "use tree-sitter repo-map instead of embeddings RAG",
          "trigger": "need code-base context",
          "decision_rule": "repo-map gives LLM symbol-level signatures + PageRank ranking; budget is dynamic (grows when no files /add-ed, shrinks when added)",
          "evidence_cite": "aider.chat/docs/repomap.html, aider.chat/2024/05/22/swe-bench-lite.html (70.3% correct-file hit)"
        },
        {
          "operation": "pick edit format by model capability, not preference",
          "trigger": "model selected",
          "decision_rule": "Aider has per-model defaults: GPT-4o/Sonnet=diff, Gemini=diff-fenced, GPT-4-Turbo=udiff, GPT-4.1=patch, weak models=whole",
          "evidence_cite": "aider.chat/docs/more/edit-formats.html"
        },
        {
          "operation": "never wrap code edits in JSON tool-calls",
          "trigger": "designing agent harness around LLM code edits",
          "decision_rule": "use plain-text diff format; JSON degrades every tested model including Sonnet",
          "evidence_cite": "aider.chat/2024/08/14/code-in-json.html"
        }
      ]
    },
    {
      "id": "R2",
      "name": "SOP Workflow",
      "operations": [
        {
          "operation": "bootstrap from clean git working tree",
          "trigger": "session start",
          "decision_rule": "cd to repo root, commit/stash in-flight work; aider will auto-commit dirty files separately if you don't",
          "evidence_cite": "aider.chat/docs/git.html"
        },
        {
          "operation": "/add only files that need editing",
          "trigger": "scoping a task",
          "decision_rule": "keep added files under 25k tokens; trust repo-map for everything else; /read for reference-only files",
          "evidence_cite": "aider.chat/docs/usage/tips.html, aider.chat/docs/troubleshooting/edit-errors.html"
        },
        {
          "operation": "discuss with /ask before /code",
          "trigger": "non-trivial change",
          "decision_rule": "use /ask to align on plan; switch to /code (or omit prefix) to execute",
          "evidence_cite": "aider.chat/docs/usage/modes.html"
        },
        {
          "operation": "use architect mode when reasoner differs from editor",
          "trigger": "best reasoner edits poorly (e.g., o1-preview)",
          "decision_rule": "pair architect (o1/o3) with capable editor (Sonnet, GPT-4o, DeepSeek); accept ~2x cost for ~3-6pp benchmark gain",
          "evidence_cite": "aider.chat/2024/09/26/architect.html"
        },
        {
          "operation": "wire --auto-lint and --test-cmd --auto-test",
          "trigger": "edit-heavy session",
          "decision_rule": "non-zero exit feeds output back to LLM which proposes fix; wrap reformatters in dual-pass script",
          "evidence_cite": "aider.chat/docs/usage/lint-test.html"
        },
        {
          "operation": "pin coding style with --read CONVENTIONS.md",
          "trigger": "team or project has consistent style preferences",
          "decision_rule": "load once via --read; example: 'prefer httpx over requests; use type hints'",
          "evidence_cite": "aider.chat/docs/usage/conventions.html"
        },
        {
          "operation": "monorepo: use --subtree-only + .aiderignore",
          "trigger": "monorepo or very large repo",
          "decision_rule": "cd into subdir, run aider with --subtree-only; exclude vendor/dist with .aiderignore",
          "evidence_cite": "aider.chat/docs/faq.html"
        },
        {
          "operation": "/clear when topic shifts; /reset for clean slate",
          "trigger": "long session drift or new task",
          "decision_rule": "/clear keeps files, drops chat; /reset drops both",
          "evidence_cite": "aider.chat/docs/usage/commands.html"
        }
      ]
    },
    {
      "id": "R3",
      "name": "Dilemma Cases",
      "operations": [
        {
          "dilemma": "model edits wrong file",
          "trigger": "LLM proposes edits to unintended/invented file path",
          "decision_rule": "/ls to inspect; /add target; /drop extras; /ask which file implements X if unknown",
          "evidence_cite": "aider.chat/docs/usage/tips.html"
        },
        {
          "dilemma": "context window overflow on large repo",
          "trigger": "/tokens > 25k; truncated responses",
          "decision_rule": "step 1 /tokens, 2 /drop, 3 /clear, 4 --map-tokens 1024 or 0, 5 --subtree-only + .aiderignore, 6 split task",
          "evidence_cite": "aider.chat/docs/troubleshooting/token-limits.html, aider.chat/docs/troubleshooting/edit-errors.html"
        },
        {
          "dilemma": "diff format hallucinates / SEARCH block not found",
          "trigger": "Aider reports 'SEARCH block not found' repeatedly",
          "decision_rule": "/clear; upgrade model; fallback --edit-format whole; try --architect for two-step reliability",
          "evidence_cite": "aider.chat/docs/troubleshooting/edit-errors.html, aider.chat/2023/12/21/unified-diffs.html"
        },
        {
          "dilemma": "is architect mode worth the cost",
          "trigger": "considering whether to pay for two LLM calls per turn",
          "decision_rule": "skip if reasoner edits cleanly; use if reasoner is lazy at editing (o1-preview 79.7%->82.7% with Sonnet); SOTA path (85%) is too slow for interactive",
          "evidence_cite": "aider.chat/2024/09/26/architect.html"
        },
        {
          "dilemma": "Sonnet truncates at 4k output tokens",
          "trigger": "Claude 3.5 Sonnet response cuts off mid-edit",
          "decision_rule": "Aider already handles via multi-response stitching + minimality prompts; update Aider; if persists, manually request minimal SEARCH/REPLACE blocks",
          "evidence_cite": "aider.chat/2024/07/01/sonnet-not-lazy.html"
        },
        {
          "dilemma": "JSON tool-call vs plain text diff for code",
          "trigger": "designing agent harness; weighing structured output",
          "decision_rule": "prefer plain-text diff; JSON degrades every tested model; if forced, single string field, expect quality hit",
          "evidence_cite": "aider.chat/2024/08/14/code-in-json.html"
        }
      ]
    },
    {
      "id": "R4",
      "name": "Anti-Patterns & Boundaries",
      "operations": [
        {
          "anti_pattern": "use Aider for greenfield with no existing codebase",
          "rationale": "repo-map empty; per-edit-commit safety has no marginal value over plain LLM chat",
          "evidence_cite": "aider.chat/docs/repomap.html"
        },
        {
          "anti_pattern": "use Aider for refactors across thousands of files",
          "rationale": "25k-token sweet spot for added files; sed/AST/IDE refactor first then Aider for semantic residual",
          "evidence_cite": "aider.chat/docs/troubleshooting/edit-errors.html"
        },
        {
          "anti_pattern": "use Aider without git",
          "rationale": "loses per-edit commit, /undo, dirty-file separation, audit trail",
          "evidence_cite": "aider.chat/docs/git.html"
        },
        {
          "anti_pattern": "use Aider for fully autonomous ticket->PR",
          "rationale": "Aider is human-in-loop by design; OpenHands/Devin target autonomy",
          "evidence_cite": "aider.chat/2024/05/22/swe-bench-lite.html"
        },
        {
          "mistake": "/add too many files for safety",
          "fix": "/drop aggressively; trust repo-map (70.3% correct-file hit on SWE-Bench Lite)",
          "evidence_cite": "aider.chat/2024/05/22/swe-bench-lite.html"
        },
        {
          "mistake": "weak model as editor in architect mode",
          "fix": "editor must follow edit format reliably; Sonnet/DeepSeek/GPT-4o are good editors",
          "evidence_cite": "aider.chat/2024/09/26/architect.html"
        },
        {
          "mistake": "skip /clear between tasks",
          "fix": "/clear when topic shifts; long history drifts model behavior",
          "evidence_cite": "aider.chat/docs/usage/tips.html"
        },
        {
          "mistake": "disable auto-commit to keep history clean",
          "fix": "leave auto-commit on; squash with git rebase -i HEAD~N at end",
          "evidence_cite": "aider.chat/docs/git.html"
        },
        {
          "mistake": "skip CONVENTIONS.md",
          "fix": "load with --read; persists style decisions across sessions",
          "evidence_cite": "aider.chat/docs/usage/conventions.html"
        }
      ]
    },
    {
      "id": "R5",
      "name": "Ecosystem Context",
      "operations": [
        {
          "comparison": "Aider vs Cline",
          "axis": "approval model",
          "decision": "Aider = retrospective /undo; Cline = prospective per-tool-call approve",
          "evidence_cite": "cline.bot/blog/top-9-cursor-alternatives-in-2025"
        },
        {
          "comparison": "Aider vs Cursor",
          "axis": "surface",
          "decision": "Aider = terminal pure + open-source; Cursor = closed IDE + visual diff overlays",
          "evidence_cite": "shakudo.io/blog/best-ai-coding-assistants"
        },
        {
          "comparison": "Aider vs Continue",
          "axis": "IDE breadth",
          "decision": "Continue = cross-IDE (VS Code + JetBrains); Aider = terminal-only",
          "evidence_cite": "frontman.sh/blog/best-open-source-ai-coding-tools-2026"
        },
        {
          "comparison": "Aider vs OpenHands",
          "axis": "autonomy",
          "decision": "Aider = human-in-loop REPL; OpenHands = autonomous sandbox ticket->PR",
          "evidence_cite": "frontman.sh/blog/best-open-source-ai-coding-tools-2026"
        },
        {
          "research_legacy": "Aider's published experiments inform any coding-agent designer",
          "items": [
            "udiff vs SEARCH/REPLACE: 20% -> 61% on GPT-4 Turbo refactor benchmark",
            "tree-sitter repo-map: 70.3% correct-file hit on SWE-Bench Lite",
            "architect+editor split: 79.7% -> 85% Polyglot pass@2",
            "JSON tool-call wrapping degrades every tested model for code editing",
            "Sonnet 4k-output workaround: 55.1% -> 64.0% refactor benchmark"
          ],
          "evidence_cite": "aider.chat/blog series 2023-2024"
        }
      ]
    }
  ]
}
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