Repository navigation
refactor(agents): Move five Sol agents to Claude Sonnet 5.5 - #758
Merged
Merged
Conversation
Requirements, Architect, IaC Planner, Challenger and Context Optimizer now use Claude Sonnet 5.5 at high effort and follow the Claude body contract. Registry, catalog, explorer graph, effort policy, skill wording and tests are updated. The utility prompts keep GPT-6.1 Sol.
Contributor
There was a problem hiding this comment.
🟡 Changes recommended
Converted Claude agents need consistent untrusted-content fencing, and the policy test does not protect all five exceptions.
4 open findings
What changed in this PR
Moves five workflow agents from GPT-6.1 Sol to Claude Sonnet 5.5 while preserving high reasoning effort.
Changes:
- Converts five agent bodies to the Claude contract.
- Synchronizes model registries and routing references.
- Updates effort-policy and model-assignment tests.
| File | Description |
|---|---|
.github/agents/02-requirements.agent.md |
Converts Requirements to Claude. |
.github/agents/03-architect.agent.md |
Converts and restructures Architect. |
.github/agents/05-iac-planner.agent.md |
Converts and streamlines IaC Planner. |
.github/agents/10-challenger.agent.md |
Converts Challenger to Claude. |
.github/agents/11-context-optimizer.agent.md |
Converts Context Optimizer to Claude. |
.github/instructions/agent-authoring.instructions.md |
Defines high-effort Sonnet exceptions. |
.github/model-catalog.json |
Updates model inventory and assignments. |
.github/skills/apex-agent-authoring/references/model-policy.md |
Documents the revised effort policy. |
.github/skills/apex-azure-defaults/references/cost-estimate-parent-contract.md |
Updates Architect routing. |
.github/skills/apex-host-workflow-start/SKILL.md |
Makes routing language model-neutral. |
.github/skills/apex-workflow-engine/references/operational-safety.md |
Generalizes specialist-step restrictions. |
.github/skills/apex-workflow-engine/references/review-imported-iac.md |
Generalizes imported-IaC routing language. |
.github/skills/apex-workflow-engine/references/workflow-entry.md |
Generalizes workflow-entry restrictions. |
tools/registry/agent-registry.json |
Mirrors new model assignments. |
tools/registry/architecture-explorer-graph.json |
Regenerates explorer metadata. |
tools/tests/scripts/test_host_adapters.mjs |
Updates routing assertion wording. |
tools/tests/scripts/test_post_write_validation.mjs |
Updates Architect model assertion. |
tools/tests/scripts/test_skill_consolidation.mjs |
Tests revised effort policy. |
tools/tests/vendor-prompting/run.test.mjs |
Tests canonical agent assignments. |
🧠 Review effort: Balanced
The tool-contracts test asserts the phrase in 03-Architect. The earlier trim dropped it.
Wrap Learn results, worker output and artifacts in pasted_content in 03, 05 and 10. Assert the full five-agent high-effort exception in the policy test.
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
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.
Description
Moves 02-Requirements, 03-Architect, 05-IaC Planner, 10-Challenger and 11-Context Optimizer from
GPT-6.1 Sol (copilot)toClaude Sonnet 5.5 (copilot). The five keepreasoning-effort: high, so the effort policy now names them as an exception to the Sonnet default ofmedium.Type of Change
Token / latency impact
This change affects input-token budget / per-call latency:
The Claude hard checkpoint (>=160K input) now applies to these five agents instead of >=300K. No measurement was taken.
Workflow Used
Changes Made
Files modified:
.github/agents/{02,03,05,10,11}-*.agent.md: model switched; bodies rewritten to the Claude contract (## Role,<scope_fencing>,<output_contract>,<stop_conditions>,<context_awareness>). 03, 05 and 11 gain<investigate_before_answering>; 02 omits it because it is one-shot. 03 and 05 were trimmed of duplicated sections to stay within the 600-line body limit.tools/registry/agent-registry.json,tools/registry/architecture-explorer-graph.json,.github/model-catalog.json: assignments updated (catalog assignments and graph regenerated by their scripts).agent-authoring.instructions.md,model-policy.md,test_skill_consolidation.mjs.apex-host-workflow-start,apex-workflow-enginereferences andcost-estimate-parent-contract.md; matching test assertions updated.Unchanged: the four utility prompts stay on GPT-6.1 Sol, so the label and the
gpt-6-solfamily remain authorized.Testing Performed
Code Quality
npm run lint:md)npm run validate:all,npm run validate:agents,npm run lint:vendor-prompting,validate-models.mjs,validate:agent-registry,generate-model-catalog.mjs --checkand the touchednode --testfiles all pass. Pre-push checks pass.Docs follow-up: a changelog entry in
jonathan-vella/apex-docs.