Reusable defaults for AI agents, skills, prompts, wrappers, examples, and benchmarkable token-efficiency workflows.
Start with the User Guide · Browse the Index · Run Validation · View Benchmarks
Modern AI engineering work needs reusable defaults: agent roles, behavior skills, task prompts, tool-specific instruction wrappers, validation checks, and benchmark artifacts.
AgentDefaults helps you:
- Bootstrap coding, DevOps, automation, research, product, documentation, or creative-production agents.
- Keep behavior consistent across repo-aware coding tools, chat models, local models, IDE assistants, and MCP-connected apps.
- Reduce input/context/output token waste without sacrificing correctness or validation.
- Benchmark concise-agent behavior instead of assuming shorter prompts are better.
- Reuse proven instruction stacks instead of rewriting the same agent setup every time.
git clone https://github.com/Quazmoz/agentdefaults.git
cd agentdefaults
python3 scripts/validate-agentdefaults.pyThen choose a path:
| I want to... | Start here |
|---|---|
| Pick the right files quickly | docs/user-guide.md |
| Use a local repo-aware coding CLI | docs/quickstarts/cli.md |
| Use Claude-style repo instructions | docs/quickstarts/claude.md |
| Use Gemini-style repo instructions | docs/quickstarts/gemini.md |
| Use editor rule files | docs/quickstarts/editor.md |
| Use repository assistant profile wrappers | docs/quickstarts/repo-assistant.md |
| Use Palmier Pro through MCP | docs/quickstarts/palmierpro-mcp.md |
| Add a new reusable default | docs/patterns/default.md |
| Copy a ready-made example | examples/ |
AgentDefaults separates canonical reusable content from thin tool wrappers.
| Layer | Folder / File | Purpose |
|---|---|---|
| Entrypoints | AGENTS.md, CLAUDE.md, GEMINI.md |
Broad repo-level instructions for supported tools. |
| Tool wrappers | .github/copilot-instructions.md, .github/agents/ |
Tool-native profiles that point back to canonical files. |
| Editor rules | .cursor/rules/agentdefaults.mdc, .windsurfrules |
Thin editor integration files. |
| Agents | agents/ |
Full reusable agent profiles. |
| Skills | skills/ |
Composable behavior and task modules. |
| Prompts | prompts/ |
Copy-paste benchmark and task prompts. |
| Docs | docs/ |
Quickstarts, integration guides, patterns, tool maps, and benchmarks. |
| Examples | examples/ |
Practical copy-paste recipes. |
| Manifest | agentdefaults.manifest.json |
Machine-readable repo summary. |
Rule: update canonical content first, then keep wrappers thin and discoverable.
Use when you want an agent to operate Palmier Pro through MCP for timeline edits, story assembly, YouTube Shorts, transcript cleanup, captions, b-roll, generation approval workflows, and exports.
docs/quickstarts/palmierpro-mcp.md
agents/palmierpro-mcp-video-editor-agent.md
skills/palmierpro-mcp-setup-and-safety.md
skills/palmierpro-timeline-editing.md
skills/palmierpro-transcript-cuts-and-captions.md
skills/palmierpro-ai-generation-workflow.md
docs/palmierpro-mcp-tool-map.md
prompts/palmierpro/story-assembly-from-project-media.md
prompts/palmierpro/youtube-short-from-long-form.md
prompts/palmierpro/full-edit-pass.md
examples/palmierpro-mcp-workflow.md
Use when a team wants to lower GitHub Copilot spend (usage-based AI Credits) without losing quality.
.github/copilot-instructions.md (drop-in from examples/copilot-token-efficiency.md)
skills/copilot-token-efficiency.md
skills/context-budgeting-and-pruning.md
skills/token-output-budgeting.md
skills/token-efficiency-measurement.md
Use when you want smaller outputs, narrower context, and measurable savings.
AGENTS.md
agents/token-economy-orchestrator.md
agents/token-efficient-response-agent.md
skills/context-budgeting-and-pruning.md
skills/token-output-budgeting.md
skills/token-efficient-response-compression.md
skills/token-efficiency-measurement.md
Use when you want focused implementation work with compact status reporting.
AGENTS.md
agents/terse-technical-coding-agent.md
skills/context-budgeting-and-pruning.md
skills/token-output-budgeting.md
skills/token-efficient-response-compression.md
Use when you want to measure baseline vs candidate behavior.
skills/token-efficiency-measurement.md
prompts/token-efficiency/common-task-benchmark.md
prompts/token-efficiency/compare-models.md
docs/benchmarks/token-efficiency-smoke-test.md
docs/benchmarks/token-efficiency-fresh-2026-06-25.md
| Tool / Runner | Primary file(s) | Notes |
|---|---|---|
| Generic repo-aware agents | AGENTS.md |
Base instruction file for broad compatibility. |
| Claude / Claude Code | CLAUDE.md |
Claude-oriented wrapper, delegates shared rules to AGENTS.md. |
| Gemini / Gemini CLI | GEMINI.md |
Gemini-oriented wrapper, delegates shared behavior to AGENTS.md. |
| GitHub Copilot repo instructions | .github/copilot-instructions.md |
Repo-wide behavior and maintenance rules. |
| GitHub Copilot custom agents | .github/agents/ |
Selectable profile wrappers. |
| Cursor | .cursor/rules/agentdefaults.mdc |
Thin rule wrapper pointing back to canonical files. |
| Windsurf | .windsurfrules |
Thin wrapper pointing back to canonical files. |
| MCP-connected apps | docs/quickstarts/palmierpro-mcp.md |
Palmier Pro video-editing workflow over local MCP. |
| Any chat/local model | agents/, skills/, prompts/ |
Copy-paste the smallest useful stack. |
For the full list, use INDEX.md.
| Recipe | Use |
|---|---|
examples/palmierpro-mcp-workflow.md |
Copy-paste Palmier Pro MCP editing workflows. |
examples/copilot-token-efficiency.md |
Drop-in .github files + habits to lower GitHub Copilot cost. |
examples/coding.md |
Build a compact coding workflow. |
examples/benchmark.md |
Run a token-efficiency benchmark. |
examples/compression.md |
Compress prompt, memory, or instruction files. |
examples/handoff.md |
Create compact continuation handoffs. |
examples/local-model.md |
Use AgentDefaults with a chat/local model. |
examples/repository-profile.md |
Use a thin repository profile wrapper. |
| Pattern | Use |
|---|---|
docs/patterns/default.md |
Generic reusable default structure. |
docs/patterns/skill.md |
New skill structure. |
docs/patterns/prompt.md |
New prompt structure. |
docs/patterns/benchmark.md |
New benchmark artifact structure. |
Run the reusable validator instead of copying long snippets from the README:
python3 scripts/validate-agentdefaults.pyIt checks:
- Required files exist.
- Markdown defaults include
## Purpose. - Local Markdown links resolve.
| Artifact | Result | Scope |
|---|---|---|
docs/benchmarks/token-efficiency-smoke-test.md |
38.8% average output savings | Initial local IDE-agent smoke test. |
docs/benchmarks/token-efficiency-fresh-2026-06-25.md |
35.4% average output savings | Fresh third-pass local benchmark after validation micro-examples. |
These are repo-internal validation artifacts, not public benchmark claims. For stronger evidence, run prompts/token-efficiency/compare-models.md across multiple model classes with exact provider token counts.
agentdefaults/
├── AGENTS.md
├── CLAUDE.md
├── GEMINI.md
├── agentdefaults.manifest.json
├── scripts/
│ └── validate-agentdefaults.py
├── docs/
│ ├── user-guide.md
│ ├── ux-roadmap.md
│ ├── palmierpro-mcp-tool-map.md
│ ├── quickstarts/
│ ├── patterns/
│ └── benchmarks/
├── examples/
├── agents/
├── skills/
├── prompts/
├── .github/
├── .cursor/
├── .windsurfrules
├── INDEX.md
└── README.md
- Compress language, not meaning.
- Reduce input/context, tool-result, and output waste.
- Preserve exact technical identifiers and safety rules.
- Benchmark quality separately from token savings.
- Keep prompts usable across hosted frontier models, coding agents, local LLMs, and MCP tools.
Before adding a new default, ask:
- Is this reusable across more than one project?
- Is the expected output clear?
- Does it define enough context for an agent to perform well?
- Are safety boundaries and destructive-action limits explicit?
- Can another engineer quickly adapt it?
- Can token-efficiency claims be measured or honestly estimated?
Use docs/patterns/default.md for new reusable content.
Usable cross-tool scaffold with benchmarked token-efficiency defaults, Palmier Pro MCP video-editing defaults, tool wrappers, quickstarts, examples, patterns, and validation tooling.
License to be added.