Repository navigation
Ground meta-orchestration subagents, remove fabricated metrics (#300) - #308
Merged
Merged
Conversation
Rewrites the knowledge-synthesizer definition to match what its tools (Read, Write, Edit, Glob, Grep) can actually do, addressing issue #300: - Removes fabricated metrics ("Pattern accuracy > 85%", "50k+ entities", "improving performance by 23%") the model would otherwise hallucinate into its output. - Drops technique blocks the tools cannot execute (federated/reinforcement learning, knowledge-graph building, anomaly/causation inference). - Replaces the circular "Communication Protocol" JSON (which assumed a non-existent message bus) with an evidence-cited output schema and an honest note that coordination happens through shared files. - Adds hard grounding rules: every pattern must cite path:line evidence and appear in >=2 sources; no invented counts. Bumps voltagent-meta 1.0.3 -> 1.0.4 (plugin.json + marketplace.json) to satisfy the version-bump CI gate. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Applies the same cleanup as knowledge-synthesizer to the rest of the
category, all of which shared the identical keyword-stuffed template:
agent-organizer, context-manager, error-coordinator,
multi-agent-coordinator, performance-monitor, task-distributor,
workflow-orchestrator
For each file:
- Removed fabricated metrics — the numeric-target checklists
(">95%", "<100ms", "100%"), the progress/status JSON blobs with
invented counts, and the "Delivery notification" brag sentences the
model would otherwise hallucinate into its output.
- Cut technique/buzzword blocks the tools (Read, Write, Edit, Glob,
Grep) cannot execute; reframed each agent honestly as planning and
documenting rather than running a live runtime.
- Replaced the circular "Communication Protocol" JSON (which assumed a
non-existent message bus) with honest file-and-orchestrator
coordination and explicit tool-scope/grounding rules.
- Preserved frontmatter (name/tools/model) and the sibling-agent
cross-references.
Each file dropped from ~286 lines to ~66-101. No version bump needed
beyond the 1.0.4 already applied for knowledge-synthesizer, since all
changes are within the voltagent-meta category.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
5 tasks done
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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
Addresses #300, which points out that the meta-orchestration agents read as walls of buzzwords with metrics and capabilities their tools cannot back up. The critique is largely fair, so this reworks the whole
09-meta-orchestrationcategory to be honest about what a Claude Code subagent withRead, Write, Edit, Glob, Grepcan actually do.All 8 files shared the identical ~286-line template. For each one:
Pattern accuracy > 85%,Retrieval time < 100ms,100%guarantees), the progress/status JSON blobs with invented counts ("patterns_identified": 342,"recovery_rate": "93%"), and the "Delivery notification" brag sentences (improving system performance by 23%,Knowledge graph contains 50k+ entities). These were pure hallucination bait — nothing computes them.path:lineevidence, don't invent numbers, flag uncertainty).name/tools/model) and the sibling-agent cross-references.Impact
Net: +419 / −2021 lines across the category.
Changes
categories/09-meta-orchestration/voltagent-meta1.0.3 → 1.0.4(plugin.json+marketplace.json) per the version-bump CI gateTest plan
name/tools/model) unchanged per file.mdchanges; marketplace in sync)🤖 Generated with Claude Code