Use small model alias in inline sub-agent blocks

.github/aw/subagents.md

@@ -34,7 +34,7 @@ Define a sub-agent with a level-2 Markdown heading of the form `## agent: \`name
 ## agent: `file-summarizer`
 ---
 description: Summarizes the content of a file in a few concise sentences
-model: claude-haiku-4.5
+model: small
 ---
 You are a file summarization assistant. When given a file path, read the
 file and return a brief summary (2–4 sentences) describing its purpose
@@ -58,7 +58,15 @@ Only two fields are supported inside a sub-agent frontmatter block:
 | Field | Required | Default | Notes |
 |---|---|---|---|
 | `description` | No | — | Human-readable summary of the sub-agent's role |
-| `model` | No | `"inherited"` | Model override; `"inherited"` uses the parent workflow's model |
+| `model` | No | `"inherited"` | Model override; `"inherited"` uses the parent workflow's model. Prefer model aliases (e.g. `small`, `large`) over specific model IDs for portability. |
+
+**Prefer model aliases over model IDs.** Built-in aliases resolve to the best available model for each provider, so they continue to work as models are updated. Commonly used aliases for sub-agents:
+
+| Alias | Resolves to | When to use |
+|---|---|---|
+| `small` | `mini` → haiku, gpt-5-mini, gpt-5-nano, gemini-flash | Cheap, fast tasks: extraction, classification, formatting |
+| `large` | sonnet, gpt-5-pro, gpt-5, gemini-pro | Complex reasoning or synthesis tasks |
+| `inherited` | Parent workflow model | Default — use when the sub-agent needs the same capability as the parent |
 
 All other fields (`engine`, `tools`, `network`, etc.) are stripped at runtime with a warning. Sub-agents inherit the parent's engine, tool access, and network configuration.
 
@@ -127,7 +135,7 @@ The top item must have a linked PR draft or issue.
 ## agent: `dependency-scanner`
 ---
 description: Lists outdated npm/pip/go packages
-model: claude-haiku-4.5
+model: small
 ---
 Run the appropriate package-manager audit command and return a
 machine-readable list of outdated packages with their current and
@@ -136,22 +144,22 @@ latest versions.
 ## agent: `test-coverage`
 ---
 description: Summarises low-coverage code paths
-model: claude-haiku-4.5
+model: small
 ---
 Read the most recent test coverage report and list the top 5 files or
 functions with coverage below 60 %. Include the file path and line range.
 
 ## agent: `secret-scanner`
 ---
 description: Checks for potential credential leaks
-model: claude-haiku-4.5
+model: small
 ---
 Scan staged changes and recently modified files for patterns that
 resemble API keys, tokens, or passwords. Report any findings with the
 file name and approximate line number.
 ```
 
-The parent model (e.g. Claude Sonnet or Copilot) orchestrates, while the sub-agents do the heavy lifting with a haiku-size model at lower cost.
+The parent model (e.g. Claude Sonnet or Copilot) orchestrates, while the sub-agents do the heavy lifting with a `small` model at lower cost.
 
 ### 2 — Reusable specialised helpers
 
@@ -184,7 +192,7 @@ tools:
 ## agent: `diff-explainer`
 ---
 description: Produces a plain-English summary of a pull request diff
-model: claude-haiku-4.5
+model: small
 ---
 You receive a unified diff. Describe each changed file in one sentence,
 focusing on *what changed* and *why it matters*. Ignore formatting-only

.github/workflows/agent-performance-analyzer.md

@@ -600,7 +600,7 @@ Execute all phases systematically and maintain an objective, data-driven approac
 
 ## agent: `metrics-extractor`
 ---
-model: claude-haiku-4.5
+model: small
 description: Reads shared repo-memory metric files and returns structured JSON with all relevant performance data
 ---
 You are a metrics extraction assistant. When given a newline-separated list of file paths (one path per line), read each file using bash and return a single JSON object containing all data found.
@@ -613,7 +613,7 @@ Return the result as a single valid JSON object with no additional commentary.
 
 ## agent: `pattern-detector`
 ---
-model: claude-haiku-4.5
+model: small
 description: Classifies agent behavioral patterns from profiles and returns a structured categorization of issues found
 ---
 You are an agent behavior classification assistant. When given a JSON object containing agent profiles (with fields such as output counts, types, success rates, and resource usage), classify each agent's behavioral patterns.

.github/workflows/daily-security-red-team.md

@@ -710,7 +710,7 @@ Initialize your cache-memory, determine today's technique, and begin your compre
 
 ## agent: `forensics-extractor`
 ---
-model: claude-haiku-4.5
+model: small
 description: Run git blame on each security finding and extract commit origin metadata as JSON
 ---
 You receive security findings as plain text, one per line, in the format:
@@ -754,7 +754,7 @@ Output one JSON object per line. No preamble, no summary.
 
 ## agent: `fix-task-generator`
 ---
-model: claude-haiku-4.5
+model: small
 description: Generate markdown remediation checklist from classified security findings
 ---
 You receive security finding records as JSON objects, one per line:

.github/workflows/daily-subagent-optimizer.md

@@ -53,7 +53,7 @@ imports:
 
 # Daily Sub-Agent Optimizer
 
-You are an LLM efficiency expert specializing in agentic workflow optimization. Your mission today: identify one workflow that would benefit from inline sub-agent refactoring, reason carefully about which tasks a smaller model (haiku/mini) can handle, and produce a copy-paste-ready proposal issue.
+You are an LLM efficiency expert specializing in agentic workflow optimization. Your mission today: identify one workflow that would benefit from inline sub-agent refactoring, reason carefully about which tasks a smaller model (small alias) can handle, and produce a copy-paste-ready proposal issue.
 
 ## Context
 
@@ -128,7 +128,7 @@ Use the `prefix-analyzer` sub-agent to perform this analysis. Pass it the full p
 
 ## Phase 5 — LLM Expert Analysis
 
-As an LLM efficiency expert, identify where the workflow's prompt does work that a smaller haiku/mini model can handle independently.
+As an LLM efficiency expert, identify where the workflow's prompt does work that a smaller model can handle independently.
 
 ### Sub-Agent Candidate Scoring
 
@@ -145,7 +145,7 @@ Threshold: **≥ 6 → strong candidate, 4–5 → moderate, < 4 → keep in mai
 
 ### Heuristics Cheatsheet
 
-Tasks a **haiku/mini model handles well**:
+Tasks a **`small` model handles well**:
 - Summarizing a single file or code section
 - Extracting specific fields from structured/semi-structured text
 - Classifying items into a predefined set of categories
@@ -171,7 +171,7 @@ Collect all sections scoring ≥ 4. Pick the **top 2–4** by score to propose a
 For each selected sub-agent candidate, design a concrete inline sub-agent:
 
 1. **Name**: lowercase, hyphenated, descriptive (e.g., `file-summarizer`, `category-detector`)
-2. **Model**: `claude-haiku-4.5`
+2. **Model**: `small`
 3. **Description**: one sentence (≤ 15 words)
 4. **Agent prompt**: focused, ≤ 15 lines, imperative mood
 5. **Invocation change**: the 1–3 line replacement in the main prompt that calls the sub-agent by name
@@ -198,7 +198,7 @@ Body:
 ### Why This Workflow
 
 [2–3 sentences: what makes it a good candidate — high token usage, number of distinct phases,
-specific tasks identified as haiku-appropriate or having repeated tool prefixes]
+specific tasks identified as small-appropriate or having repeated tool prefixes]
 
 ---
 
@@ -242,11 +242,11 @@ Then remove the duplicate calls from each of the affected sections.
 
 ### Proposed Sub-Agents
 
-#### 1. `<agent-name>` (`claude-haiku-4.5`)
+#### 1. `<agent-name>` (`small`)
 
 **Extracted task**: [1 sentence]
-**Why haiku**: [1 sentence — which heuristic applies]
-**Score**: <X>/10 (independence: N, haiku-adequacy: N, parallelism: N, size: N)
+**Why small**: [1 sentence — which heuristic applies]
+**Score**: <X>/10 (independence: N, model-adequacy: N, parallelism: N, size: N)
 **Estimated savings**: ~N tokens/run
 
 <details>
@@ -256,7 +256,7 @@ Then remove the duplicate calls from each of the affected sections.
 ## agent: `<agent-name>`
 ---
 description: <description>
-model: claude-haiku-4.5
+model: small
 ---
 <agent prompt>
 ```
@@ -340,7 +340,7 @@ Load the existing array from that path if the file is present, append the new en
 ## agent: `prefix-analyzer`
 ---
 description: Detects repeated tool-call prefixes across workflow prompt sections and scores extraction value
-model: claude-haiku-4.5
+model: small
 ---
 You are a prompt-structure analyst. Given the full body text of an agentic workflow prompt (everything after the closing frontmatter `---`), identify repeated tool invocations that appear as opening instructions in multiple sections.
 
@@ -380,7 +380,7 @@ reasoning: <1–2 sentences explaining the finding>
 ## agent: `workflow-screener`
 ---
 description: Reads a workflow .md file and reports whether inline-agents are enabled, the engine, and prompt complexity
-model: claude-haiku-4.5
+model: small
 ---
 You are a workflow file scanner. When given a file path, read the file using bash and report the following facts:
 
@@ -403,27 +403,27 @@ notes: <one sentence>
 
 ## agent: `opportunity-classifier`
 ---
-description: Scores a workflow prompt section on its suitability for extraction into a haiku/mini sub-agent
-model: claude-haiku-4.5
+description: Scores a workflow prompt section on its suitability for extraction into a small sub-agent
+model: small
 ---
-You are an LLM task-decomposition expert. Given a section of an agentic workflow prompt, score it on its suitability to be extracted into a sub-agent using a smaller haiku/mini model.
+You are an LLM task-decomposition expert. Given a section of an agentic workflow prompt, score it on its suitability to be extracted into a sub-agent using a smaller model.
 
 Score each dimension:
 
 - **independence** (0–3): Can this section run without the outputs of other sections? 3 = fully independent, 0 = deeply coupled to earlier results
-- **haiku_adequacy** (0–3): Is the reasoning simple enough for a smaller model? 3 = pure extraction/classification/formatting, 0 = requires deep synthesis or cross-referencing many sources
+- **model_adequacy** (0–3): Is the reasoning simple enough for a smaller model? 3 = pure extraction/classification/formatting, 0 = requires deep synthesis or cross-referencing many sources
 - **parallelism** (0–2): Could this run concurrently with other sections? 2 = yes, 0 = must be sequential
 - **size** (0–2): Is the task substantial enough to warrant a separate agent call? 2 = many tool calls or long output, 0 = trivial (< 2 tool calls)
 
-Compute: `total = independence + haiku_adequacy + parallelism + size` (max 10)
+Compute: `total = independence + model_adequacy + parallelism + size` (max 10)
 
 Verdict: `strong` (≥ 6), `moderate` (4–5), `weak` (< 4)
 
 Return in this exact format:
 ```
 total: <score>/10
 independence: <0-3>
-haiku_adequacy: <0-3>
+model_adequacy: <0-3>
 parallelism: <0-2>
 size: <0-2>
 verdict: strong/moderate/weak

.github/workflows/developer-docs-consolidator.md

@@ -537,7 +537,7 @@ Begin the consolidation process now. Use Serena for analysis, **directly apply c
 ## agent: `file-cataloger`
 ---
 description: Discover and catalog all markdown files in specs/ and scratchpad/
-model: claude-haiku-4.5
+model: small
 ---
 You receive no arguments. Discover all markdown files in the `specs/` and `scratchpad/` directories using bash:
 
@@ -561,7 +561,7 @@ Return only the table, no other commentary.
 ## agent: `tone-analyzer`
 ---
 description: Scan a markdown file for marketing language and formatting violations
-model: claude-haiku-4.5
+model: small
 ---
 You receive a single file path as your input. Read the file and perform two scans:
 

.github/workflows/smoke-copilot.md

@@ -210,7 +210,7 @@ If all tests pass and this workflow was triggered by a pull_request event:
 
 ## agent: `file-summarizer`
 ---
-model: claude-haiku-4.5
+model: small
 description: Summarizes the content of a file in a few concise sentences
 ---
 You are a file summarization assistant. When given a file path, read the file and return a brief summary (2–4 sentences) describing its purpose and key contents. Be concise and factual.

.github/workflows/spec-librarian.md

@@ -264,7 +264,7 @@ The following specifications are outdated:
 
 ## agent: `coverage-checker`
 ---
-model: claude-haiku-4.5
+model: small
 description: Lists all pkg/ packages and reports README.md coverage metrics as JSON
 ---
 You are a coverage auditor for a Go repository. Your task is to enumerate all packages
@@ -292,7 +292,7 @@ Return ONLY a JSON object with these six fields and no additional text.
 
 ## agent: `staleness-detector`
 ---
-model: claude-haiku-4.5
+model: small
 description: Compares git timestamps for each package's source vs spec and detects API drift
 ---
 You are a staleness detector for Go package specifications. You receive a list of packages
@@ -330,7 +330,7 @@ Return ONLY the JSON object and no additional text.
 
 ## agent: `consistency-checker`
 ---
-model: claude-haiku-4.5
+model: small
 description: Validates import paths, naming conventions, and dependency declarations across all specs
 ---
 You are a cross-package consistency checker for Go package specifications. Your task is to

.github/workflows/unbloat-docs.md

@@ -502,7 +502,7 @@ Begin by scanning the docs directory and selecting the best candidate for improv
 
 ## agent: `file-bloat-analyzer`
 ---
-model: claude-haiku-4.5
+model: small
 description: Reads a single documentation file and returns a structured inventory of bloat indicators
 ---
 You are a documentation bloat analysis agent. The file path to analyze is provided as the first line of your input (or as the argument you are invoked with). Read that file using the `bash` tool (`cat <file_path>`) and return a structured JSON inventory of bloat indicators.
@@ -533,7 +533,7 @@ Return a JSON object only — no prose, no extra text:
 
 ## agent: `doc-page-screenshotter`
 ---
-model: claude-haiku-4.5
+model: small
 description: Navigates to a documentation page URL using Playwright and captures a full-page screenshot, returning a structured JSON result with screenshot paths and any blocked domains
 ---
 You are a documentation screenshot agent. Your input is a full page URL to screenshot (e.g., `http://localhost:4321/gh-aw/guides/ephemerals/`).