feat: add ce:ideate skill with issue-grounded ideation mode

[tmchow]

The problem

When you ask an AI coding agent "what should I build next?" or "what would you improve?", you typically get generic, surface-level suggestions — things like "add more tests" or "improve documentation." The ideas aren't grounded in your actual codebase, they aren't critically filtered, and there's no structured way to capture them or act on them.

The compound-engineering plugin already has ce:brainstorm (define what a chosen idea means) and ce:plan (figure out how to build it). But there's a missing first step: figuring out which ideas are even worth exploring. Many times, you have a roadmap already but other times, it can be helpful to get an ideation thought partner.

What this PR adds

ce:ideate — a new skill for grounded idea generation

ce:ideate answers the question: "What are the strongest ideas worth exploring in this project?"

It works by:

1. Scanning the actual codebase to understand project shape, patterns, pain points, and leverage points — so every idea is grounded in reality, not abstract advice
2. Generating 30-40 raw ideas using parallel sub-agents (7-8 per agent), each approaching from a different creative frame (user friction, missing capabilities, automation opportunities, assumption-breaking, compounding leverage, edge cases)
3. Adversarially filtering the full list — skeptic agents attack the candidates, and only ideas that survive explicit rejection criteria make it through. Every rejected idea gets a reason.
4. Presenting 5-7 survivors with a scannable overview line, then detailed blocks with downsides and confidence scores
5. Writing a durable artifact to docs/ideation/ so the work survives session interruptions and can be resumed later

The skill routes into ce:brainstorm (not straight to implementation), completing the workflow: ideate → brainstorm → plan → work.

Issue-grounded ideation mode

When a user's argument indicates they want issue-tracker data as input (e.g., /ce:ideate bugs, /ce:ideate github issues), ce:ideate activates a new issue intelligence agent alongside its existing codebase scan:

• New issue-intelligence-analyst agent — fetches GitHub issues via gh CLI (with GitHub MCP as fallback), clusters them by root-cause themes (not individual symptoms), and returns structured analysis with trend direction, confidence scores, and source mix (human vs bot-generated)
• Priority-aware fetching — scans for priority/severity labels and fetches high-priority issues first, so P0s aren't buried under recent P3s
• Token-efficient — truncated bodies (500 chars) included in the initial bulk fetch via --jq, avoiding N+1 individual body reads
• Hybrid frame strategy — issue clusters drive ideation frames, padded with default frames when fewer than 4 themes exist. This keeps ideas grounded in real user pain while maintaining creative diversity
• Closed issues as recurrence signal — closed issues don't create themes on their own (that's a solved problem). They only strengthen themes that already have open issues — the recurrence of fixed-then-reopened patterns is the real smell that something systemic isn't resolved
• Standalone capable — the agent works independently outside ce:ideate for general issue landscape analysis

Example: running /ce:ideate bugs on a repo with 25+ duplicate LIVE_DOC_UNAVAILABLE issues produces a "collaboration write path reliability" theme — not 25 separate bug-fix ideas.

Other improvements

• Volume model tuning — reduced per-agent idea target from 10 to 7-8 based on real usage data showing the speculative tail rarely survived filtering
• Rejection summary as numbered table — easier to scan than a flat bullet list, and makes filtering rigor tangible

Files changed

File	Change
agents/research/issue-intelligence-analyst.md	New agent
skills/ce-ideate/SKILL.md	Issue-tracker intent detection, conditional agent dispatch, dynamic frame derivation, resume awareness, rejection table format
plugin.json	Agent count 28 → 29
marketplace.json	Description updated
README.md	Research agents table updated
docs/brainstorms/	Requirements doc for the feature
docs/plans/	Implementation plan

[kieranklaassen]

idea: I love this one. Could we make the compoundEngineeringConfig also define what the sources are you want to use normally in the ideation? For example, I can link posthog, AppSignal, maybe Postgres database, stuff like that. So you can customize what the sources for ideation are normally.

[kieranklaassen]

but please merge and then i can also add these things