Designer Drift Report — 2026-08-03
Summary
2 drift items found across 2 categories: missing frontmatter field coverage (skills, lsp, evals) and a missing cross-reference to the newly added Ad Hoc Evaluation Mode / Failure Classification content. Trigger, safe-output, network-ecosystem, and pattern coverage are all fully in sync.
View Drift Items by Category
Frontmatter Field Coverage
- Source:
.github/aw/syntax.md (index → skills.md, lsp.md, evals.md)
- Missing from:
.github/aw/designer.md (both the Decision Heuristics mapping tables and the Generation Template)
- Details:
syntax.md documents three top-level frontmatter fields — skills (skill loading), lsp (language server integration), and evals (BinEval binary evaluations) — but designer.md never mentions any of them in its "User says... → Maps to" tables, its Generation Template skeleton, or its References section.
- Suggested fix: Add a short mapping row for each (e.g. "uses reusable skill bundles" →
skills:, "wants LSP-backed code intelligence" → lsp:, "wants binary pass/fail evaluations" → evals:), and add skills.md/lsp.md/evals.md to the References list alongside the other syntax sub-files.
Recent Changes Without Sync
- Source:
.github/aw/create-agentic-workflow.md (Ad Hoc Evaluation Mode, Multi-Scenario Evaluation Example, Failure Classification — added in commit 5325893)
- Missing from:
.github/aw/designer.md and .github/agents/interactive-agent-designer.agent.md
- Details: The same commit that added
designer.md and the interactive-agent-designer agent also introduced a distinct "Ad Hoc Evaluation Mode" workflow (exploratory recommendations without file creation, multi-scenario comparison tables, and a 3-tier failure classification: invocation_unavailable / command_not_supported / response_unavailable) in create-agentic-workflow.md. Neither designer.md nor the interactive-agent-designer agent references this mode, so a user asking the designer for an exploratory "what would you recommend" evaluation has no pointer to it.
- Suggested fix: Add a short note in
designer.md's "When to Use This Skill" section pointing to create-agentic-workflow.md#ad-hoc-evaluation-mode for exploratory/no-file-creation requests, mirroring the existing cross-references to create-agentic-workflow.md and agentic-chat.md.
No Drift
- ✅ Trigger coverage — all trigger types in
triggers.md (pull_request, schedule/fuzzy scheduling, slash_command, label_command, workflow_dispatch, deployment_status, workflow_run, issues, forks) are represented in designer.md's Trigger Mapping table.
- ✅ Safe output coverage — all ~44 safe-output action types from
safe-outputs*.md and all-safe-output-types.txt are represented in designer.md's Safe Output Mapping table.
- ✅ Network/tool coverage — all real ecosystem identifiers from
network.md (the "Valid Values" and "Ecosystem Identifiers" tables) appear in designer.md's Network Mapping table; the "missing" entries found during comparison (npm, pip, cargo, docker, etc.) are network.md's documented invalid shorthands, not real identifiers, so this is not drift.
- ✅ Pattern coverage — all 19 named patterns from
patterns.md (MonitorOps, BatchOps, ChatOps, DailyOps, IssueOps, etc.) are referenced in designer.md's Pattern Heuristics table.
- ✅ Stale references — no docs, fields, or options referenced by the designer files were found to be removed from the reference docs.
Methodology
Compared outlines of 8 reference docs (syntax.md, safe-outputs.md, network.md, patterns.md, subagents.md, token-optimization.md, triggers.md, create-agentic-workflow.md) against both designer files (.github/aw/designer.md and .github/agents/interactive-agent-designer.agent.md).
Reference doc commits (last 7 days): 1
Designer file commits (last 7 days): 1 (same commit — designer.md and the interactive-agent-designer agent were added together with the reference docs, but the Ad Hoc Evaluation Mode content added in create-agentic-workflow.md in that same commit was not cross-linked from the designer files)
Generated by 🔍 Designer Drift Audit · auto · 81.5 AIC · ⌖ 2.5 AIC · ⊞ 7.4K · ◷
Designer Drift Report — 2026-08-03
Summary
2 drift items found across 2 categories: missing frontmatter field coverage (
skills,lsp,evals) and a missing cross-reference to the newly added Ad Hoc Evaluation Mode / Failure Classification content. Trigger, safe-output, network-ecosystem, and pattern coverage are all fully in sync.View Drift Items by Category
Frontmatter Field Coverage
.github/aw/syntax.md(index →skills.md,lsp.md,evals.md).github/aw/designer.md(both the Decision Heuristics mapping tables and the Generation Template)syntax.mddocuments three top-level frontmatter fields —skills(skill loading),lsp(language server integration), andevals(BinEval binary evaluations) — butdesigner.mdnever mentions any of them in its "User says... → Maps to" tables, its Generation Template skeleton, or its References section.skills:, "wants LSP-backed code intelligence" →lsp:, "wants binary pass/fail evaluations" →evals:), and addskills.md/lsp.md/evals.mdto the References list alongside the other syntax sub-files.Recent Changes Without Sync
.github/aw/create-agentic-workflow.md(Ad Hoc Evaluation Mode, Multi-Scenario Evaluation Example, Failure Classification — added in commit 5325893).github/aw/designer.mdand.github/agents/interactive-agent-designer.agent.mddesigner.mdand the interactive-agent-designer agent also introduced a distinct "Ad Hoc Evaluation Mode" workflow (exploratory recommendations without file creation, multi-scenario comparison tables, and a 3-tier failure classification:invocation_unavailable/command_not_supported/response_unavailable) increate-agentic-workflow.md. Neitherdesigner.mdnor the interactive-agent-designer agent references this mode, so a user asking the designer for an exploratory "what would you recommend" evaluation has no pointer to it.designer.md's "When to Use This Skill" section pointing tocreate-agentic-workflow.md#ad-hoc-evaluation-modefor exploratory/no-file-creation requests, mirroring the existing cross-references tocreate-agentic-workflow.mdandagentic-chat.md.No Drift
triggers.md(pull_request, schedule/fuzzy scheduling, slash_command, label_command, workflow_dispatch, deployment_status, workflow_run, issues, forks) are represented indesigner.md's Trigger Mapping table.safe-outputs*.mdandall-safe-output-types.txtare represented indesigner.md's Safe Output Mapping table.network.md(the "Valid Values" and "Ecosystem Identifiers" tables) appear indesigner.md's Network Mapping table; the "missing" entries found during comparison (npm,pip,cargo,docker, etc.) arenetwork.md's documented invalid shorthands, not real identifiers, so this is not drift.patterns.md(MonitorOps, BatchOps, ChatOps, DailyOps, IssueOps, etc.) are referenced indesigner.md's Pattern Heuristics table.Methodology
Compared outlines of 8 reference docs (
syntax.md,safe-outputs.md,network.md,patterns.md,subagents.md,token-optimization.md,triggers.md,create-agentic-workflow.md) against both designer files (.github/aw/designer.mdand.github/agents/interactive-agent-designer.agent.md).Reference doc commits (last 7 days): 1
Designer file commits (last 7 days): 1 (same commit —
designer.mdand the interactive-agent-designer agent were added together with the reference docs, but the Ad Hoc Evaluation Mode content added increate-agentic-workflow.mdin that same commit was not cross-linked from the designer files)