Evaluate and synthesize research findings from ALL agents, help the user select features, and update the main research document. This transitions research from in-progress to in-evaluation (matching the feature pipeline).

Argument Resolution

If no ID is provided, or the ID doesn't match an existing topic in progress or in-evaluation:

List all files in ./docs/specs/research-topics/03-in-progress/ and ./docs/specs/research-topics/04-in-evaluation/ matching research-*.md
If a partial ID or name was given, filter to matches
Present the matching topics and ask the user to choose one

Recommended: Use a Different Model

For unbiased evaluation, use a different model than those that conducted the research.

claude --model sonnet
/aigon-research-eval 05

Step 1: Run the CLI command

IMPORTANT: You MUST run this command first. It transitions the engine state to evaluating and refreshes the generated lifecycle view.

aigon research-eval $1

Step 2: Read All Findings

Find and read ALL findings files:

docs/specs/research-topics/logs/research-{ID}-*-findings.md

Step 3: Synthesize Findings

Present to the user:

Consensus

What all agents agree on.

Divergent Views

Where agents disagree and why - be specific about which agent said what.

Step 4: Consolidate Features

Extract the ## Suggested Features table from each agent's findings file.

Deduplication rules:

Exact match: Same feature name from multiple agents = one entry, note all agents
Similar concept: Different names but same idea = merge into one, pick the best name
Related but distinct: Keep separate but note the relationship
Unique: One agent only = keep, but flag for user attention

Present a consolidated table:

## Consolidated Features

| # | Feature Name | Description | Priority | Agents | Status |
|---|--------------|-------------|----------|--------|--------|
| 1 | feature-name | Best description | high | cc, gg | Consensus |
| 2 | another-feat | Description | medium | cc | Unique to Claude |
| 3 | third-feat | Description | low | gg, cx | Consensus |

Status values:

Consensus - Multiple agents suggested (stronger signal)
Unique to [Agent] - Only one agent suggested
Merged - Combined similar suggestions from multiple agents

Step 5: Get User Approval and Create Features

Before pausing, signal the dashboard that you are blocked on user input so the card gets a pulsing "Awaiting input" badge and the user gets a desktop notification (they may not be watching the tmux pane):

aigon agent-status awaiting-input "Pick which of the consolidated features to create. Reply with numbers, 'all', 'consensus', or 'none'."

Then ask the user:

"Here are the consolidated features. Which should I create?

Enter numbers to include (e.g., 1,2,3)

Enter all to include everything

Enter consensus to include only consensus items

Enter none to skip feature creation"

Wait for user response before proceeding. The awaiting-input flag clears automatically on the next agent-status write.

Feature Set Naming

Default: group as a set. Features created from the same research topic share context and are almost always related. Unless the user explicitly opts out, stamp every spec with a shared set: slug and wire dependencies between them.

Step A — Derive the plan (do this before asking anything)

Derive a set slug:

lowercase the topic slug
trim the leading research-<id>- prefix if present
keep it short and descriptive (example: research-34-feature-set → feature-set)

Name features with common prefix + sequence numbers in dependency order (feature 1 has no deps, feature 2 depends on 1, etc.):

<prefix>-1-<specific-name>
<prefix>-2-<specific-name>
<prefix>-3-<specific-name>

Analyse the dependency chain across the selected features. For each feature, decide: does it depend on another selected feature? A feature depends on another when it cannot be safely started until that one is merged (shared data model, API contract, etc.). Be concrete — "feature 2 needs the xterm.js bundle from feature 1" — not vague.

Step B — Present the full plan and wait (REQUIRED CHECKPOINT)

Do not run any feature-create command until the user responds to this prompt.

Present the complete plan in one message:

"I’ll group these as set <slug>. Here’s the naming and dependency plan:

# Name Depends on
1 <prefix>-1-<name> —
2 <prefix>-2-<name> #1
3 <prefix>-3-<name> #2

Dependencies are based on: [brief reason — e.g., "feature 2 needs the WS layer from feature 1"].

Group these as set <slug>? (y/n/edit slug)

Enter or y = confirm (creates with --set <slug> and wires depends_on)

n = create without set grouping or dependencies

edit slug = adjust the slug before confirming

Adjust numbering/deps inline if the order is wrong"

Step C — Create features

If confirmed (default):

aigon feature-create "feature-name" --set <slug>
# repeat for each feature, e.g.:
aigon feature-create "prefix-1-name" --set <slug>
aigon feature-create "prefix-2-name" --set <slug>

If declined:

aigon feature-create "feature-name"

After creating each feature, edit the spec to add:

research: frontmatter field (REQUIRED — stamps a structured link from feature → research):

---
complexity: medium
research: $1
---

The dashboard reads this field to populate the FEATURES sub-tab inside the research detail panel. Use the numeric research ID from this command's argument; do not invent or quote it.

Research origin backlink in the body:

## Related
- Research: #{ID} {research-name}

For features that depend on another feature in this set, add depends_on to the Dependencies section:

## Dependencies
- depends_on: prefix-1-name

This enables Aigon’s dependency system to enforce ordering — dependent features cannot be started until their dependencies are done. Every feature that has a predecessor must have this field; omitting it is a mistake.

After all specs are edited, immediately prioritise every feature in the set in dependency order (roots first, leaves last). This assigns numeric IDs in execution order, so the board reflects the correct sequence at a glance.

aigon feature-prioritise "prefix-1-name"   # root — no deps
aigon feature-prioritise "prefix-2-name"   # depends on #1
aigon feature-prioritise "prefix-3-name"   # depends on #2

Never prioritise in creation order or alphabetical order. Always follow the dependency chain: if feature B depends on feature A, A must be prioritised first so it gets the lower ID.

Step 6: Update Main Research Doc

Once user confirms, update the main research document:

Update ## Recommendation with your synthesized recommendation.

Update ## Output with the selected features:

## Output

### Set Decision

- Proposed Set Slug: `feature-set`
- Chosen Set Slug: `feature-set`
<!-- If declined, write: Chosen Set Slug: none (declined) -->

### Selected Features

| Feature Name | Description | Priority | Create Command |
|--------------|-------------|----------|----------------|
| feature-name | Description | high | `aigon feature-create "feature-name" --set feature-set` |

### Feature Dependencies
<!-- List dependency chains so features can be prioritised in order -->
<!-- Each feature spec should have depends_on in its Dependencies section -->
- feature-b depends on feature-a (feature-b spec has `depends_on: feature-a`)

### Not Selected
<!-- Features discussed but not selected, for reference -->
- other-feature: Reason not selected

If the user declined grouping, keep the Create Command column without --set and record Chosen Set Slug: none (declined) so a later re-evaluation can preserve that decision explicitly.

Step 7: Commit and hand off

THIS IS THE FINAL STEP. YOU MUST COMPLETE IT. DO NOT SKIP THIS STEP.

After updating the document, commit your changes:

git add docs/specs/research-topics/ docs/specs/features/01-inbox/
git commit -m "docs: research evaluation for $1"

Then tell the user:

"Evaluation complete. Selected features have been created with research backlinks. Run /aigon:research-close {ID} when ready."

Do not run aigon agent-status research-complete — that signal is for the per-agent findings phase (03-in-progress). Closing out the evaluation is a user decision; they run research-close when they're satisfied with the output.

STAY in the session. The user may want to review the evaluation or ask for changes.

Important

Read ALL findings files, not just your own
Be objective when presenting - don't favor your own findings
Wait for user confirmation before updating files or creating features
Use the exact table format so output is clean and actionable
Created features must include a research origin backlink
Every created feature spec MUST have research: $1 in its frontmatter — this is what powers the dashboard FEATURES sub-tab

ARGUMENTS: $ARGUMENTS

#	Name	Depends on
1	`<prefix>-1-<name>`	—
2	`<prefix>-2-<name>`	#1
3	`<prefix>-3-<name>`	#2