synthesize

name	synthesize
description	Use when multiple experiments or analyses have accumulated and their findings need to be interpreted together
complexity	opus-only
model-minimum	opus
disable-model-invocation	false
allowed-tools	["Read","Grep","Glob"]
argument-hint	[project name, time range, topic, or file paths]

/synthesize

You are synthesizing accumulated findings to surface patterns, contradictions, and insights that individual log entries or analyses miss on their own. The argument specifies the scope: a project name, a time range, a topic, or specific file paths.

Pre-flight audit

Before writing synthesis output, run the synthesis pre-flight audit (docs/sops/synthesis-preflight-audit.md). Enumerate upstream sources, flag provisional data, and spot-check key numerical claims that will be cited. This prevents the most common synthesis failure: propagating contaminated or stale numbers from prior sessions.

Gather material

Based on the scope argument:

• If a project name: read the project README (especially Log and Open questions), any files in the project directory, and related decision records.
• If a time range: scan logs across all active projects for entries in that range.
• If a topic: grep across projects, decisions, and docs for relevant material.
• If file paths: read those files directly.

Also check decisions/ for relevant recorded choices and docs/ for framework documents.

Analyze across CI layers

For the gathered material, identify:

1. Cross-layer causal chains — Findings that connect across CI layers. (e.g., "The evaluation gap [L4] exists because the interface [L3] can't present 3D interactively to LLMs, which limits what the model [L1] can judge.")
2. Convergent signals — Multiple independent findings pointing to the same conclusion. What do they converge on?
3. Contradictions — Findings that conflict with each other. Which is better grounded? What would resolve the disagreement?
4. Gaps — What questions remain unasked? What CI layers are underrepresented in the findings? What experiments would fill the gaps?
5. Gravity candidates — Recurring patterns that should be formalized. What manual work could become automated tooling? What tooling could become model capability?

Output format

## Synthesis: <scope>

### Material reviewed
<bulleted list of files/entries consulted>

### Cross-layer chains
<numbered findings, each tracing a connection across 2+ CI layers>

### Convergent signals
<what multiple findings agree on — with specific references>

### Contradictions
<conflicting findings and what would resolve them>

### Gaps
<what's missing — specific questions or unexamined CI layers>

### Gravity candidates
<patterns that should move downward — from manual to tool to model>

### Implications
<1-3 concrete recommendations for what to do next, referencing specific projects or actions>

Prioritize insight density over comprehensiveness. A synthesis that surfaces one genuine cross-layer insight is more valuable than one that restates what the logs already say.

Save to disk

Write the synthesis to projects/<project>/analysis/<scope-slug>-synthesis-YYYY-MM-DD.md. Use the project most relevant to the synthesis scope.

Task Bridge

After saving the synthesis, convert actionable implications to tasks:

1. For each item in the "Implications" section that contains a concrete action verb (implement, create, run, update, design, investigate, fix):
   • Check the project's TASKS.md for an existing task covering the same action
   • If no existing task, create one:
     • Imperative description derived from the implication
     • [fleet-eligible] or [requires-opus] per fleet-eligibility checklist
     • [skill: ...] tag matching the work type
     • Done when: from the implication's expected outcome
     • Why: referencing this synthesis file path
2. For "Gaps" that suggest specific experiments or investigations: create tasks referencing /design or /diagnose
3. For "Gravity candidates" rated "formalize now": create a task to run /gravity on the candidate
4. Skip implications that are purely observational or contextual

Cross-session synthesis insights are among the highest-value outputs the system produces. Converting them to tasks ensures they are acted upon rather than rediscovered.

Commit

Follow docs/sops/commit-workflow.md. Commit message: synthesize: <scope summary>