feat: improve agent swarm with better descriptions and dynamic awareness #6
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Changed from spec-only consultant to full implementation agent: - Mode: subagent → all (core implementation agent) - Permissions: edit and bash now allowed (with ask) - Chrome DevTools: enabled for visual inspection and verification Workflow: inspect → understand patterns → implement → verify visually Maintains bold aesthetic philosophy (no generic AI look) but now writes actual CSS, component styling, layouts, and motion code instead of specifications.
Expanded role to help other agents when stuck on problems: Consultation Mode (new): - Analyzes bug reports and error messages - Suggests debugging strategies with specific steps - Provides root cause analysis with confidence levels - Guides without implementing (advisory only) - Examples: executor stuck on bug, tester can't diagnose failure Design Mode (existing): - Solution design with options and recommendations - Architecture guidance for new features - Trade-off analysis Temperature increased 0.3 → 0.5 for creative problem-solving. Updated orchestrator to route stuck agents to architect.
Added {{agents}}, {{agents:table}}, and {{agents:list}} template variables
that expand to formatted lists of available agents.
Features:
- Filters out disabled agents (AgentConfig.disabled === true)
- Only includes agents with descriptions
- Truncates descriptions to 80 chars with ellipsis
- Sorted alphabetically by agent ID
- Backwards compatible (ctx parameter optional)
Formats:
- {{agents}} or {{agents:table}} → Markdown table
- {{agents:list}} → Bullet list with bold agent names
Updated all 11 agent index.ts files to pass context to expandVariables.
Orchestrator prompt now uses {{agents:table}} instead of hardcoded list.
Fixed ordering issue where agents only saw agents set up before them.
Solution: Two-phase setup
1. Phase 1: All agents set up with raw prompts (no expansion)
2. Phase 2: expandAgentPrompts() expands all prompts after registration
This ensures {{agents}} references see the complete agent list regardless
of setup order.
Descriptions now focus on ONE JOB + modes/levels:
- Removed delegation information from all descriptions
- Added constraints where applicable (READ-ONLY, ADVISORY-ONLY, IDEATION-ONLY)
- Kept concise (1-2 sentences max)
Added {{agents:table}} to all agent prompts:
- Every agent now knows the full swarm
- Can delegate to any agent via Task tool
- Placed after role description in each prompt
This enables true swarm collaboration - any agent can delegate
to any other agent based on the dynamically expanded agents table.
- Enhanced all agent descriptions with 'Use when:' guidance for better delegation - Added missing description to compaction agent - Streamlined prompts by extracting shared protocols into reusable templates - Removed tester agent (consolidated into executor/reviewer workflows) - Refactored hook files: hooks.ts → hook.ts naming convention - Consolidated task tools with proper types - Simplified protocol templates (context-handling, error-handling, escalation, plan-versioning)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
{{agents:list}}and{{agents:table}}placeholders, so agents always see the current swarm configurationChanges
expandAgentsList()andexpandAgentsTable()utilities for prompt templateshooks.ts→hook.tsconventionTesting
bun run typecheck✅bun run lint✅bun run build✅