wanshuiyin/meta-optimize

Meta-Optimize: Outer-Loop Harness Optimization for ARIS

Analyze accumulated usage logs and propose optimizations for: $ARGUMENTS

Context

ARIS is a research harness — a system of skills, bridges, workflows, and artifact contracts that wraps around LLMs to orchestrate research. This skill implements a prototype outer loop that observes how the harness is used and proposes improvements to the harness itself (not to the research artifacts it produces).

Inspired by Meta-Harness (Lee et al., 2026): the key insight is that harness design matters as much as model weights, and harness engineering can be partially automated by logging execution traces and using them to guide improvements.

What This Skill Optimizes (Harness Components)

| Component | Example | Optimizable? | |-----------|---------|:---:| | SKILL.md prompts | Reviewer instructions, quality gates, step descriptions | Yes | | Default parameters | difficulty: medium, MAX_ROUNDS: 4, threshold: 6/10 | Yes | | Convergence rules | When to stop the review loop, retry counts | Yes | | Workflow ordering | Skill chain sequence within a workflow | Yes | | Artifact schemas | What fields go in EXPERIMENT_LOG.md, idea-stage/IDEA_REPORT.md | Cautious | | MCP bridge config | Which reviewer model, routing rules | No (infra) |

Not optimized: The research artifacts themselves (papers, code, experiments). That's what the regular workflows do.

Prerequisites

1. Logging must be active. Codex mirror installs do not create Claude Code hooks. Provide .aris/meta/events.jsonl from a Codex-compatible event logger, an external wrapper, or a manually exported trace log before running this skill.
2. Sufficient data. At least 5 complete workflow runs logged in .aris/meta/events.jsonl. The skill will check and warn if insufficient.

Workflow

Step 0: Check Data Availability

EVENTS_FILE=".aris/meta/events.jsonl"
if [ ! -f "$EVENTS_FILE" ]; then
    echo "ERROR: No event log found at $EVENTS_FILE"
    echo "Enable Codex-compatible logging first: create .aris/meta/events.jsonl from your Codex wrapper, external event logger, or exported trace log."
    exit 1
fi

EVENT_COUNT=$(wc -l < "$EVENTS_FILE")
SKILL_INVOCATIONS=$(grep -c '"skill_invoke"' "$EVENTS_FILE" || echo 0)
SESSIONS=$(grep -c '"session_start"' "$EVENTS_FILE" || echo 0)

echo "📊 Event log: $EVENT_COUNT events, $SKILL_INVOCATIONS skill invocations, $SESSIONS sessions"

if [ "$SKILL_INVOCATIONS" -lt 5 ]; then
    echo "⚠️  Insufficient data (<5 skill invocations). Continue using ARIS normally and re-run later."
    exit 0
fi

Step 1: Analyze Usage Patterns

Read .aris/meta/events.jsonl and compute:

Frequency analysis:

• Which skills are invoked most often?
• Which slash commands do users type most?
• What parameter overrides are most common? (These suggest bad defaults.)

Failure analysis:

• Which tools fail most often? In which skills?
• What error patterns repeat? (OOM, import, compilation, timeout)
• How many auto-debug retries per workflow run?

Convergence analysis (for auto-review-loop):

• Average rounds to reach threshold
• Score trajectory shape (fast improvement? plateau? oscillation?)
• Which review round catches the most critical issues?
• Do users override difficulty mid-run?

Human intervention analysis:

• Where do users interrupt with manual prompts during workflows?
• What manual corrections do users make most? (These indicate skill gaps.)

Model-delta analysis (harness diet):

• Has the session model or the pinned reviewer model changed since a skill's SKILL.md was last touched? A model bump is a trigger to re-read, not evidence by itself: a deletion proposal must cite TARGET-SPECIFIC evidence (a capability-specific release note, or repeated post-bump event-log behavior showing the scaffold is unused). Never deletion candidates: privilege boundaries, acceptance/review gates, corpus/provenance rules, output contracts, safety checks. The diet targets model-compensation scaffolding only — a capability the new model has natively is pure overhead. A harness that only ever grows is a harness nobody is re-reading.

Trigger-rate analysis (measured, not from the event log): the log shows which skills were USED, not which were WANTED-but-omitted. Mainline ships tools/meta_opt/trigger_eval.py, which measures Claude Code's skill triggering via claude -p probes (trigger / confusion / miss). It is Claude-Code-specific — there is no equivalent codex skill-selection probe yet, so for a Codex executor treat trigger-rate as a mainline signal, not a step you run here. Measure-only regardless: a low rate is INPUT to a proposal, never a self-applied rewrite; the confusion matrix (which sibling a query lands on) points at disambiguation, not "make it pushier".

Present findings as a structured summary table.

Step 1.5: Name the Current Bottleneck

Synthesize the Step-1 analyses into one sentence naming the single most-limiting pipeline stage right now — e.g. "planning", "verification quality", "experiment execution reliability", "writing polish" — with evidence. The bottleneck always moves: coding → planning → verification → taste. Step 2's ranked table should read as sub-fixes for this one named constraint.

Append the verdict to the append-only ledger .aris/meta/bottleneck_log.jsonl (never edit or delete prior lines — succession history is the point):

mkdir -p .aris/meta
# json.dumps, NOT hand-interpolated shell strings: bottleneck/evidence are
# natural language — a stray quote must not break the JSONL (or the shell).
python3 - <<'PY'
import json, datetime
entry = {
    "ts": datetime.datetime.now().astimezone().isoformat(timespec="seconds"),
    "cycle": 3,
    "bottleneck": "verification quality",
    "evidence": "review rounds plateau at 6/10 while tool failures are rare",
    "top_patch_ids": ["P1", "P2"],
}
with open(".aris/meta/bottleneck_log.jsonl", "a", encoding="utf-8") as fh:
    fh.write(json.dumps(entry, ensure_ascii=False) + "\n")
PY

On the next run, read the last line first and open the report by stating whether that bottleneck was resolved and what it has moved to.

Step 2: Identify Optimization Targets

Based on Step 1, rank optimization opportunities by expected impact:

## Optimization Opportunities (ranked)

| # | Target | Signal | Proposed Change | Expected Impact |
|---|--------|--------|-----------------|-----------------|
| 1 | auto-review-loop default threshold | Users override to 7/10 in 60% of runs | Change default from 6/10 to 7/10 | Fewer manual overrides |
| 2 | experiment-bridge retry count | 40% of runs hit max retries on OOM | Add OOM-specific recovery (reduce batch size) | Fewer failed experiments |
| 3 | paper-write de-AI patterns | Users manually fix "delve" in 80% of runs | Add "delve" to default watchword list | Fewer manual edits |
| 4 | experiment-bridge Phase-2 hand-holding steps | Model bump; scaffold untouched for 2 generations; zero failures in the guarded steps | **DELETE steps N–M — the new model does this unprompted** | Smaller harness, less drift surface |

The Proposed-Change column is explicitly allowed to be a deletion — "DELETE step N, new model does this for free" is a first-class optimization.

If $ARGUMENTS specifies a target skill, focus analysis on that skill only. If $ARGUMENTS is empty or "all", analyze all skills with sufficient data.

Step 3: Generate Patch Proposals

For each optimization target, generate a concrete diff:

--- a/skills/auto-review-loop/SKILL.md
+++ b/skills/auto-review-loop/SKILL.md
@@ -15,7 +15,7 @@
 ## Constants
 
-- **SCORE_THRESHOLD = 6** — Minimum review score to accept.
+- **SCORE_THRESHOLD = 7** — Minimum review score to accept. (Raised based on usage data: 60% of users overrode to 7+.)

Rules for patch generation:

• One patch per optimization target
• Each patch must include a comment explaining WHY (with data from the log)
• Patches must be minimal — change only what the data supports
• Never change artifact schemas or MCP bridge config in v1
• Never change behavior that would break existing user workflows
• Anti-self-poisoning screen: resolve $ARIS_REPO from .aris/installed-skills-codex.txt, then run $ARIS_REPO/tools/capture_filter.py (or project-local tools/capture_filter.py) against each proposed patch rationale. If it flags an environment failure, transient error, negative tool-capability claim, or one-off narrative, rewrite the proposal to the fix/config/workaround or drop it. Warn-and-skip only when the helper cannot be resolved. See capture-antipatterns.md.

Step 4: Fresh-Agent Review of Patches (same-family provisional)

Send each patch to GPT-5.6-Sol xhigh for adversarial review:

spawn_agent:
  model: gpt-5.6-sol
  reasoning_effort: xhigh
  message: |
    You are reviewing a proposed optimization to an ARIS SKILL.md file.
    
    ## Original Skill (relevant section)
    [paste original]
    
    ## Proposed Patch
    [paste diff]
    
    ## Evidence from Usage Log
    [paste summary stats]
    
    Review this patch:
    1. Does the evidence support the change?
    2. Could this change hurt other use cases?
    3. Is the change minimal and safe?
    4. Score 1-10: should this be applied?
    
    If score < 7, explain what additional evidence would be needed.

Step 5: Present Results

Output a structured report:

# ARIS Meta-Optimization Report

**Date**: [today]
**Data**: [N] events, [M] skill invocations, [K] sessions
**Target**: [skill name or "all"]

## Current Bottleneck

**[one-phrase name]** — [one-line evidence]. Prior cycle's bottleneck: [name —
resolved by <patch ids> / unresolved / first recorded cycle]. (Ledger:
`.aris/meta/bottleneck_log.jsonl`)

## Proposed Changes

### Change 1: [title]
- **Target**: [skill/file:line]
- **Signal**: [what the data shows]
- **Patch**: [diff]
- **Reviewer Score**: [X/10]
- **Reviewer Notes**: [summary]
- **Status**: ✅ Recommended / ⚠️ Needs more data / ❌ Rejected

### Change 2: ...

## Changes NOT Made (insufficient evidence)
- [pattern observed but too few samples]

## Recommendations
- [ ] Apply Change 1 (reviewer approved)
- [ ] Collect more data for Change 3 (need N more runs)
- [ ] Consider manual review of Change 2

## Next Steps
Run `/meta-optimize apply 1` to apply a specific change, or
`/meta-optimize apply all` to apply all recommended changes.

Step 6: Apply Changes (if user approves)

If user runs /meta-optimize apply [N]:

1. Back up original SKILL.md to .aris/meta/backups/
2. Apply the patch
3. Log the change to .aris/meta/optimizations.jsonl
4. Remind user to test the changed skill on their next run

Never auto-apply without user approval.

Key Rules

• Log-driven, not speculative. Every proposed change must cite specific data from the event log. No "I think this would be better."
• Minimal patches. Change one thing at a time. Don't rewrite entire skills — the one sanctioned large edit is a scaffolding deletion backed by TARGET-SPECIFIC model-delta evidence (capability-specific release note, or repeated post-bump event-log behavior; a model-name change alone is never sufficient). Privilege boundaries, acceptance gates, corpus/provenance rules, output contracts, and safety checks are never deletion candidates. Deletions go through the same review + approval gates.
• Reviewer-gated. Every patch goes through fresh-agent same-family provisional review before recommendation.
• Reversible. Always back up before applying. Always log what changed.
• User-approved. Never auto-apply. Present, explain, let the user decide.
• Honest about uncertainty. If the data is insufficient, say so. Don't optimize on noise.
• Portable. Optimizations should improve the skill for all users, not just one user's style. If a change seems user-specific, flag it.

Event Schema Reference

The log at .aris/meta/events.jsonl contains JSONL records with these shapes:

{"ts":"...","session":"...","event":"skill_invoke","skill":"auto-review-loop","args":"difficulty: hard"}
{"ts":"...","session":"...","event":"PostToolUse","tool":"Bash","input_summary":"pdflatex main.tex"}
{"ts":"...","session":"...","event":"spawn_agent","tool":"spawn_agent","input_summary":"review..."}
{"ts":"...","session":"...","event":"tool_failure","tool":"Bash","input_summary":"python train.py"}
{"ts":"...","session":"...","event":"slash_command","command":"/auto-review-loop","args":""}
{"ts":"...","session":"...","event":"user_prompt","prompt_preview":"change difficulty to hard"}
{"ts":"...","session":"...","event":"session_start","source":"startup","model":"claude-opus-4-6"}
{"ts":"...","session":"...","event":"session_end"}

Triggering

This skill is NOT part of the standard W1→W1.5→W2→W3→W4 pipeline. It is a maintenance workflow with three trigger mechanisms:

1. Passive logging (always on): Claude Code hooks record events to .aris/meta/events.jsonl automatically during normal usage. Zero user effort.

2. Automatic readiness check (SessionEnd hook): When a Claude Code session ends, check_ready.sh counts skill invocations since the last /meta-optimize run. If ≥5 new invocations have accumulated, it prints a reminder:

   📊 ARIS has logged 8 skill runs since last optimization. Run /meta-optimize to check for improvement opportunities.
   

   This is a suggestion only — it does not auto-run optimization.

3. Manual trigger: User runs /meta-optimize when they see the reminder or whenever they want.

After each /meta-optimize run, the skill writes the current timestamp to .aris/meta/.last_optimize so the readiness check only counts new invocations.

Acknowledgements

Inspired by Meta-Harness (Lee et al., 2026) — end-to-end optimization of model harnesses via filesystem-based experience access and agentic code search.

Output Protocols

Follow these shared protocols for all output files:

• Output Versioning Protocol — write timestamped file first, then copy to fixed name
• Output Manifest Protocol — log every output to MANIFEST.md
• Output Language Protocol — respect the project's language setting

Review Tracing

After each reviewer agent call, save the trace following shared-references/review-tracing.md (Policy C — forensic; never silently skip). Use save_trace.sh (resolved per the chain in shared-references/integration-contract.md §2) or write files directly to .aris/traces/<skill>/<date>_run<NN>/. Respect the --- trace: parameter (default: full).