---
name: inference-format-optimizer
description: >-
  Iterative benchmarking, evaluation, and algorithmic optimization of alternative A2UI inference formats (such as Express, Atom, and Elemental). Trigger when asked to: (1) Run optimization passes or loops on an inference format, (2) Evaluate or benchmark format accuracy, latency, or token efficiency, (3) Create parallel worktree subagents for format iteration, or (4) Benchmark format trade-offs against baselines.
metadata:
  internal: true
title: inference-format-optimizer
canonical_url: https://skilld.dev/gh/a2ui-project/a2ui/inference-format-optimizer
last_updated: 2026-11-12T06:16:11.000Z
---

> Index of every page on this site: [llms.txt](https://skilld.dev/llms.txt)

> **Skill from skilld.dev.** Follow the user's requested action.
> If the user asked to fork this Skill, follow the fork workflow below. Do not execute the Skill instructions while copying it.
> Otherwise, follow the instructions below for this session. You do not need to install anything.
>
> Supporting files, fetch one when the Skill refers to it: [references/agent_instructions.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/references/agent_instructions.md), [references/inference_format_iteration.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/references/inference_format_iteration.md), [references/scoring_model.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/references/scoring_model.md), [references/subagent_protocol.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/references/subagent_protocol.md), [scripts/compare_results.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/compare_results.py), [scripts/optimize_format.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/optimize_format.py), [scripts/README.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/README.md), [scripts/sync_history.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/sync_history.py), [scripts/utils/__init__.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/utils/__init__.py), [scripts/utils/archiver.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/utils/archiver.py), [scripts/utils/format_tools.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/utils/format_tools.py), [scripts/utils/reporter.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/utils/reporter.py), [scripts/utils/runner.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/scripts/utils/runner.py), [templates/subagent_prompt.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/templates/subagent_prompt.md), [tests/test_compare_results.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/tests/test_compare_results.py), [tests/test_optimize_format.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/tests/test_optimize_format.py), [tests/test_sync_history.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/tests/test_sync_history.py), [tests/test_utils.py](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/tests/test_utils.py).
>
> If the user asked to install this Skill, run `npx skilld install a2ui-project/a2ui/inference-format-optimizer`. Install writes the Skill files into the project, so every session loads them.
>
> ## Fork workflow
>
> A fork creates an editable local Skill with its original author and licence. The request authorizes copying and local installation.
> 1. Check `./skills/inference-format-optimizer`, the project lockfile, and selected Agent targets together. If the local directory or installed Skill exists, stop. Never overwrite an existing directory or Agent target.
> 2. Read [source metadata](https://skilld.dev/api/v1/skills/a2ui-project/a2ui/inference-format-optimizer) once. Use sourceUrl, sourceCommit, skillPath, sourceGone, and license. If the source is gone or its path is missing, stop. If license is null, read licence files at the source commit.
> 3. Fetch only the source commit into a temporary Git repository. Do not clone full history. Derive repository_url from sourceUrl, including repository renames. If sourceCommit is absent, resolve the sourceUrl ref once. Set source_commit to that actual commit. Run these commands in one shell call:
>
> ```sh
> git init --quiet "$temporary_dir"
> git -C "$temporary_dir" fetch --quiet --depth=1 "$repository_url" "$source_commit"
> git -C "$temporary_dir" checkout --quiet --detach FETCH_HEAD
> ```
>
> Read applicable licence declarations and notices at that commit. If copying is not permitted, report the restriction and stop.
> 4. Inspect source entries together, then copy the directory containing skillPath into `./skills/inference-format-optimizer`. Use the user's path if selected. Keep the original SKILL.md, relative links, scripts, binary assets, and executable modes. Exclude .git metadata. Reject symlinks and paths outside the Skill directory. After checking entries, use cp -a where available. A regular source directory needs no custom copy script. Do not save this page wrapper as SKILL.md.
> Preserve author credit, notices, and applicable licence files from repository or parent directories. Add PROVENANCE.md with the Skill page, source URL, actual commit, original path, and licence. Retain any existing PROVENANCE.md and record new provenance separately. Batch source inspection, copying, and provenance work where practical.
> 5. In the project root, run `skilld install ./skills/inference-format-optimizer --mode copy --plain`. If skilld is unavailable, use `npx skilld install ./skills/inference-format-optimizer --mode copy --plain`. This known command needs no help lookup. Install does not support --json. Use detected Agent targets, or add --agent for the targets the user selected. Install the local path, never the upstream selector. If installation fails, preserve the local copy and report the exact failure.
> 6. Confirm the local lockfile source and installed Agent copies once. Report the local path, actual commit, and Agent targets. After edits, reinstall the same local path. Upstream updates must not replace it. Do not publish or push unless the user asks.

# Inference Format Optimizer

This skill provides procedural workflows, CLI orchestrators, decision guardrails, and subagent protocols for iteratively optimizing A2UI inference formats.

---

## Quick-Start CLI Cheatsheet

All execution scripts live under `scripts/` in this skill:

| Action                          | Executable Command                                                                                                                                                                |
| :------------------------------ | :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Run Fast Validation Eval**    | `python scripts/optimize_format.py --format <format>`                                                                                                                             |
| **Run Full Evaluation Suite**   | `python scripts/optimize_format.py --format <format> --full`                                                                                                                      |
| **Test Parsing / Compilation**  | `python scripts/optimize_format.py --format <format> --compile "(Card (Text \"Hi\"))"`                                                                                            |
| **Compare vs Baseline**         | `python scripts/compare_results.py --baseline eval/iterative_format_optimizer/baselines/<format>/unbounded_run_meta.json eval/iterative_format_optimizer/logs/temp_optimization/` |
| **Archive Run Artifacts**       | `python scripts/optimize_format.py --format <format> --archive --hypothesis "..." --status KEEP [--history-dir <path>]`                                                           |
| **Sync Multi-Worktree History** | `python scripts/sync_history.py [--history-dir <path>]`                                                                                                                           |

---

## Detailed References

- **Scoring & Decision Rules**: See [references/scoring_model.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/references/scoring_model.md) for $S_{\text{opt}}$ formula, correctness guardrails, and efficiency caps.
- **Subagent Worktree Protocol**: See [references/subagent_protocol.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/references/subagent_protocol.md) for launching subagents in isolated git worktrees.
- **Subagent Prompt Template**: See [templates/subagent_prompt.md](https://skilld.dev/api/skills-raw/a2ui-project/a2ui/inference-format-optimizer/templates/subagent_prompt.md) for launching pass tasks via `invoke_subagent`.

---

## The 6-Step Optimization Workflow

1. **Analyze History**: Inspect past runs in `eval/iterative_format_optimizer/history/<format>/` and read `eval/iterative_format_optimizer/history_summary.md` to avoid repeating past reverted hypotheses.
2. **Implement Hypothesis**: Modify `compiler.py`, `prompt_generator.py`, or `parser.py` under the format's package: `python/a2ui_agent/src/a2ui/inference_formats/express/` for Express, or `python/a2ui_agent/src/a2ui/inference_formats/experimental/<format>/` for an experimental format.
3. **Run Unit Conformance Tests**: Verify code changes pass pytest unit tests.
4. **Execute Benchmark Evaluation**: Run `python scripts/optimize_format.py --format <format>`.
5. **Evaluate Decision Rules**:
   - Must pass Pytest and maintain baseline accuracy.
   - Code Output Tokens must NOT expand $> +5\%$.
   - Keep change if composite score $S_{\text{opt}}$ improves; revert otherwise (`git reset --hard HEAD`).
6. **Archive & Synchronize**: Archive run with `--archive` and update history index using `python scripts/sync_history.py`.
