feat(models): offer the Claude 5 family only, default to Sonnet 5

.github/README.md

@@ -235,14 +235,14 @@ Empty `var` = the skill's default behavior (scan everything, auto-pick topics).
 The default model for all skills is set in `aeon.yml` (or from the dashboard header dropdown):
 
 ```yaml
-model: claude-sonnet-4-6
+model: claude-sonnet-5
 ```
 
-Options: `claude-sonnet-4-6` (default), `claude-opus-4-8`, `claude-fable-5`, `claude-opus-4-7`, `claude-sonnet-5`, `claude-haiku-4-5-20251001`. Per-run overrides are available via workflow dispatch, and individual skills can override to optimize cost:
+Options: `claude-sonnet-5` (default), `claude-opus-5`, `claude-haiku-4-5-20251001`. Per-run overrides are available via workflow dispatch, and individual skills can override to optimize cost:
 
 ```yaml
 skills:
-  token-movers: { enabled: true, schedule: "30 12 * * *", model: "claude-sonnet-4-6" }
+  token-movers: { enabled: true, schedule: "30 12 * * *", model: "claude-haiku-4-5-20251001" }
 ```
 
 > These ids apply to the **claude** harness. Each other [harness](../docs/harnesses.md) carries its own model list (Codex `openai/*`, Kimi `moonshotai/*`, Vibe `mistralai/*`, Pi `deepseek/*`), which the dashboard picker swaps in when you select that harness.

.github/workflows/aeon.yml

@@ -23,11 +23,8 @@ on:
         # "Resolve harness" forwards them verbatim as the harness model.
         options:
           - '(config default)'
-          - claude-opus-4-8
-          - claude-opus-4-7
-          - claude-fable-5
           - claude-sonnet-5
-          - claude-sonnet-4-6
+          - claude-opus-5
           - claude-haiku-4-5-20251001
           - grok-4.5
           - openai/gpt-5-mini
@@ -287,7 +284,7 @@ jobs:
           #   3. a per-harness cheap default.
           # aeon-native ids (claude-*/grok-*) mean nothing to an OpenRouter CLI, so
           # they're treated as "unset" and fall through to the default — a repo that
-          # never touched the model picker (still `model: claude-sonnet-4-6`) still
+          # never touched the model picker (still `model: claude-sonnet-5`) still
           # gets a working default instead of a dead id. The greps carry `|| true`
           # for the same reason as the harness greps above (no key ≠ fatal).
           # Each harness defaults to its own native family (the dashboard's
@@ -782,7 +779,7 @@ jobs:
           elif [ -n "$SKILL_MODEL" ]; then
             MODEL="$SKILL_MODEL"
           else
-            MODEL="${CONFIG_MODEL:-claude-sonnet-4-6}"
+            MODEL="${CONFIG_MODEL:-claude-sonnet-5}"
           fi
 
           # --- Harness resolution (which agent CLI runs the skill) ---

.github/workflows/messages.yml

@@ -445,7 +445,7 @@ jobs:
         run: |
           TODAY=$(date -u +%Y-%m-%d)
           CONFIG_MODEL=$(grep -E '^model:' aeon.yml | sed 's/^model: *//' | tr -d ' ')
-          MODEL="${CONFIG_MODEL:-claude-sonnet-4-6}"
+          MODEL="${CONFIG_MODEL:-claude-sonnet-5}"
 
           # Harness resolution — inbound messaging isn't a per-skill run, so it
           # follows the repo's global `harness:` in aeon.yml (same default/guard as

aeon.yml

@@ -2,7 +2,7 @@
 #
 # Enable skills by setting enabled: true. Multiple skills at the same time run in parallel.
 # Set var: "value" to pass a default parameter to the skill.
-# Set model: "claude-sonnet-4-6" per skill to override the default model (cost optimization).
+# Set model: "claude-haiku-4-5-20251001" per skill to override the default model (cost optimization).
 # Set harness: "grok" per skill to run it through the Grok Build CLI instead of the
 # default Claude Code harness (see the top-level `harness:` key below).
 
@@ -177,14 +177,16 @@ chains:
   #     - skill: routine, consume: [token-movers, hn-digest, issue-triage, github-trending]
 
 # Default model for all skills. Override per-run via workflow_dispatch.
-# Options: claude-opus-4-8, claude-fable-5, claude-sonnet-5, claude-sonnet-4-6, claude-haiku-4-5-20251001
-# (Grok harness models: grok-4.5 — flagship reasoning model & grok's current default,
-#  run with --no-subagents in CI; grok-composer-2.5-fast — fast/cheap single-agent.
-#  The older grok-build-0.1 is gateway-only, via XAI_API_KEY + the GROK_MODEL var.)
+# Options: claude-sonnet-5, claude-opus-5, claude-haiku-4-5-20251001
+# (Grok harness model: grok-4.5, the flagship reasoning model and grok's current
+#  default, run with --no-subagents in CI. It's the only id the grok CLI accepts on an
+#  X-account login, and it's the gateway default too. The older coding ids
+#  (grok-build-0.1, grok-composer-2.5-fast, grok-4.3) are gateway-only, reachable via
+#  XAI_API_KEY + the GROK_MODEL var.)
 # The non-claude harnesses use their OWN model ids (codex openai/*, kimi moonshotai/*,
 # vibe mistralai/*, pi deepseek/*) — surfaced in the dashboard picker per harness; see
 # docs/harnesses.md. The value below only applies to the claude/grok harnesses.
-model: claude-sonnet-4-6
+model: claude-sonnet-5
 
 # Agent harness — which coding-agent CLI runs your skills.
 #   claude (default)  — Claude Code (`claude -p`); uses the AI Gateway below.

apps/dashboard/app/page.tsx

@@ -42,7 +42,7 @@ export default function Dashboard() {
   const [skills, setSkills] = useState<Skill[]>([])
   const [runs, setRuns] = useState<Run[]>([])
   const [secrets, setSecrets] = useState<Secret[]>([])
-  const [model, setModel] = useState('claude-sonnet-4-6')
+  const [model, setModel] = useState('claude-sonnet-5')
   const [harness, setHarness] = useState<Harness>('claude')
   const [gateway, setGateway] = useState<GatewayProvider>('auto')
   const [repo, setRepo] = useState('')

apps/dashboard/lib/config.test.ts

@@ -26,16 +26,16 @@ import {
 const MINIMAL_YAML = `skills:
   heartbeat: { enabled: true, schedule: "0 12 * * *" }
 
-model: claude-sonnet-4-6
+model: claude-sonnet-5
 `;
 
 const FULL_YAML = `# Aeon configuration
 skills:
   morning-brief: { enabled: false, schedule: "0 7 * * *" }
-  market-pulse: { enabled: true, schedule: "0 12 * * *", model: "claude-sonnet-4-6" }
+  market-pulse: { enabled: true, schedule: "0 12 * * *", model: "claude-sonnet-5" }
   heartbeat: { enabled: true, schedule: "0 12 * * *" }
 
-model: claude-opus-4-8
+model: claude-opus-5
 
 gateway:
   provider: direct
@@ -55,7 +55,7 @@ describe("parseConfig", () => {
     assert.equal(config.skills["heartbeat"].schedule, "0 12 * * *");
     assert.equal(config.skills["heartbeat"].var, "");
     assert.equal(config.skills["heartbeat"].model, "");
-    assert.equal(config.model, "claude-sonnet-4-6");
+    assert.equal(config.model, "claude-sonnet-5");
   });
 
   it("parses a full config with all fields", () => {
@@ -64,16 +64,16 @@ describe("parseConfig", () => {
     assert.equal(config.skills["morning-brief"].enabled, false);
     assert.equal(config.skills["morning-brief"].schedule, "0 7 * * *");
     assert.equal(config.skills["market-pulse"].enabled, true);
-    assert.equal(config.skills["market-pulse"].model, "claude-sonnet-4-6");
-    assert.equal(config.model, "claude-opus-4-8");
+    assert.equal(config.skills["market-pulse"].model, "claude-sonnet-5");
+    assert.equal(config.model, "claude-opus-5");
     assert.equal(config.gateway.provider, "direct");
     assert.equal(config.jsonrenderEnabled, true);
   });
 
-  it("defaults model to claude-sonnet-4-6 when absent", () => {
+  it("defaults model to claude-sonnet-5 when absent", () => {
     const yaml = `skills:\n  test: { enabled: false, schedule: "0 0 * * *" }\n`;
     const config = parseConfig(yaml);
-    assert.equal(config.model, "claude-sonnet-4-6");
+    assert.equal(config.model, "claude-sonnet-5");
   });
 
   it("defaults gateway to auto when absent", () => {
@@ -106,7 +106,7 @@ describe("parseConfig", () => {
 
   it("handles empty skills section", () => {
     // yaml library parses `skills:` with no entries as null, not an empty map
-    const yaml = `skills:\n\nmodel: claude-sonnet-4-6\n`;
+    const yaml = `skills:\n\nmodel: claude-sonnet-5\n`;
     const config = parseConfig(yaml);
     assert.equal(Object.keys(config.skills).length, 0);
   });
@@ -138,8 +138,8 @@ describe("updateSkillInConfig", () => {
   });
 
   it("sets a model override on a skill", () => {
-    const updated = updateSkillInConfig(MINIMAL_YAML, "heartbeat", { model: "claude-sonnet-4-6" });
-    assert.ok(updated.includes("model: claude-sonnet-4-6") || updated.includes("model: 'claude-sonnet-4-6'"));
+    const updated = updateSkillInConfig(MINIMAL_YAML, "heartbeat", { model: "claude-sonnet-5" });
+    assert.ok(updated.includes("model: claude-sonnet-5") || updated.includes("model: 'claude-sonnet-5'"));
   });
 
   it("returns original yaml for non-existent skill", () => {
@@ -164,9 +164,9 @@ describe("updateSkillInConfig", () => {
 
 describe("updateModelInConfig", () => {
   it("updates the top-level model", () => {
-    const updated = updateModelInConfig(MINIMAL_YAML, "claude-opus-4-8");
+    const updated = updateModelInConfig(MINIMAL_YAML, "claude-opus-5");
     const config = parseConfig(updated);
-    assert.equal(config.model, "claude-opus-4-8");
+    assert.equal(config.model, "claude-opus-5");
   });
 
   it("replaces an existing model", () => {
@@ -321,7 +321,7 @@ describe("addSkillToConfig", () => {
   });
 
   it("inserts at end when heartbeat is absent", () => {
-    const yaml = `skills:\n  alpha: { enabled: false, schedule: "0 0 * * *" }\n\nmodel: claude-sonnet-4-6\n`;
+    const yaml = `skills:\n  alpha: { enabled: false, schedule: "0 0 * * *" }\n\nmodel: claude-sonnet-5\n`;
     const updated = addSkillToConfig(yaml, "beta");
     const config = parseConfig(updated);
     assert.ok(config.skills["beta"]);
@@ -353,9 +353,9 @@ describe("round-trip config mutations", () => {
   });
 
   it("update model and gateway independently", () => {
-    let yaml = updateModelInConfig(MINIMAL_YAML, "claude-opus-4-8");
+    let yaml = updateModelInConfig(MINIMAL_YAML, "claude-opus-5");
     const config1 = parseConfig(yaml);
-    assert.equal(config1.model, "claude-opus-4-8");
+    assert.equal(config1.model, "claude-opus-5");
 
     // Model change should not affect skills
     assert.equal(config1.skills["heartbeat"].enabled, true);
@@ -390,7 +390,7 @@ describe("upsertSkillInConfig", () => {
     const config = parseConfig(yaml);
     assert.equal(config.skills["market-pulse"].schedule, "0 6 * * *");
     // Pre-existing fields survive, and the key appears exactly once.
-    assert.equal(config.skills["market-pulse"].model, "claude-sonnet-4-6");
+    assert.equal(config.skills["market-pulse"].model, "claude-sonnet-5");
     assert.equal(yaml.match(/^\s*market-pulse:/gm)?.length, 1);
   });
 

apps/dashboard/lib/config.ts

@@ -47,7 +47,7 @@ export function parseConfig(raw: string): AeonConfig {
     }
   }
 
-  const model = String(doc.get('model') ?? 'claude-sonnet-4-6')
+  const model = String(doc.get('model') ?? 'claude-sonnet-5')
   const harnessRaw = String(doc.get('harness') ?? 'claude')
   const harness: Harness = HARNESSES.find(h => h === harnessRaw) ?? 'claude'
 

apps/dashboard/lib/constants.ts

@@ -5,11 +5,8 @@ import { HARNESS_AUTH } from './harness-auth'
 // harness-switch snap uses (modelsForHarness(...)[0] in app/page.tsx). Keep it in
 // sync with the config default in lib/config.ts and aeon.yml `model:`.
 export const MODELS = [
-  { id: 'claude-sonnet-4-6', label: 'Sonnet 4.6' },
-  { id: 'claude-opus-4-8', label: 'Opus 4.8' },
-  { id: 'claude-opus-4-7', label: 'Opus 4.7' },
-  { id: 'claude-fable-5', label: 'Fable 5' },
   { id: 'claude-sonnet-5', label: 'Sonnet 5' },
+  { id: 'claude-opus-5', label: 'Opus 5' },
   { id: 'claude-haiku-4-5-20251001', label: 'Haiku 4.5' },
 ]
 

docs/ADK.md

@@ -189,7 +189,7 @@ Everything on-demand goes through one workflow - `aeon.yml` accepts `workflow_di
 |---|---|
 | `skill` | required; must match `^[a-zA-Z0-9_-]+$` and a directory under `skills/` |
 | `var` | the skill's input for this run |
-| `model` | override, e.g. `claude-opus-4-8` - must be one of the workflow's `choice` options or GitHub rejects with **422** |
+| `model` | override, e.g. `claude-opus-5` - must be one of the workflow's `choice` options or GitHub rejects with **422** |
 | `harness` | `claude` (default) or `grok` - see [Harnesses](harnesses.md) |
 
 ```ts

docs/attestation.md

@@ -205,7 +205,7 @@ To read the richer metadata, verify the manifest and inspect its bytes:
 ```bash
 gh attestation verify output/.attest/<skill>-<run_id>.json --repo <owner>/<repo>
 cat output/.attest/<skill>-<run_id>.json
-# { "skill": "...", "model": "claude-opus-4-8", "mode": "read-only",
+# { "skill": "...", "model": "claude-opus-5", "mode": "read-only",
 #   "trigger": "schedule", "commit": "<sha>", "run_id": "<id>",
 #   "output": { "path": "output/.attest/<skill>-<run_id>.md", "sha256": "<digest>" } }
 ```

harness-adapter/adapters/pi.sh

@@ -27,7 +27,7 @@ command -v pi >/dev/null 2>&1 || {
 ARGS=(--mode json --no-session --approve)
 
 # model: pi is multi-provider — pass anything through (its registry matches
-# patterns like "claude-sonnet-4-6", "openai/gpt-4o", or bare "sonnet")
+# patterns like "claude-sonnet-5", "openai/gpt-4o", or bare "sonnet")
 [ -n "${RH_MODEL:-}" ] && [ "${RH_MODEL}" != "default" ] && ARGS+=(--model "$RH_MODEL")
 
 # read-only -> tool subsetting (pi's only native lever; advisory without the

scripts/llm-gateway.sh

@@ -183,14 +183,14 @@ case "${GATEWAY:-direct}" in
     echo "::notice::Routing through Bankr Gateway (https://llm.bankr.bot)"
     ;;
 
-  openrouter)  # NATIVE — Anthropic "skin", carries Opus 4.8
+  openrouter)  # NATIVE — Anthropic "skin", carries Opus 5
     require_secret OPENROUTER_API_KEY
     export ANTHROPIC_BASE_URL="https://openrouter.ai/api"   # NOT /api/v1
     export ANTHROPIC_AUTH_TOKEN="$OPENROUTER_API_KEY"       # Bearer; API_KEY must be blank
     unset ANTHROPIC_API_KEY CLAUDE_CODE_OAUTH_TOKEN
     # Map EVERY model slot Claude Code uses to OpenRouter slugs (opus/sonnet/haiku).
-    export ANTHROPIC_DEFAULT_OPUS_MODEL="${OPENROUTER_MODEL:-anthropic/claude-opus-4.8}"
-    export ANTHROPIC_DEFAULT_SONNET_MODEL="${OPENROUTER_MODEL_SONNET:-anthropic/claude-sonnet-4.6}"
+    export ANTHROPIC_DEFAULT_OPUS_MODEL="${OPENROUTER_MODEL:-anthropic/claude-opus-5}"
+    export ANTHROPIC_DEFAULT_SONNET_MODEL="${OPENROUTER_MODEL_SONNET:-anthropic/claude-sonnet-5}"
     export ANTHROPIC_DEFAULT_HAIKU_MODEL="${OPENROUTER_MODEL_HAIKU:-anthropic/claude-haiku-4.5}"
     MODEL="$ANTHROPIC_DEFAULT_OPUS_MODEL"
     echo "::notice::Routing through OpenRouter (Anthropic-native) as ${MODEL}"
@@ -201,7 +201,7 @@ case "${GATEWAY:-direct}" in
     export ANTHROPIC_BASE_URL="https://api.usepod.ai/proxy/${USEPOD_TOKEN}"
     export ANTHROPIC_AUTH_TOKEN="unused"    # UsePod auths via the path token
     unset ANTHROPIC_API_KEY CLAUDE_CODE_OAUTH_TOKEN
-    # UsePod mirrors the upstream Anthropic surface, so aeon's claude-opus-4-8 id
+    # UsePod mirrors the upstream Anthropic surface, so aeon's claude-opus-5 id
     # is passed through by default. If UsePod needs marketplace-specific ids, set
     # USEPOD_MODEL (+ _SONNET / _HAIKU) to override.
     if [ -n "${USEPOD_MODEL:-}" ]; then MODEL="$USEPOD_MODEL"; fi
@@ -218,10 +218,13 @@ case "${GATEWAY:-direct}" in
     export ANTHROPIC_BASE_URL="${XAI_ANTHROPIC_BASE_URL:-https://api.x.ai}"
     export ANTHROPIC_AUTH_TOKEN="$XAI_API_KEY"   # Bearer; API_KEY must be blank
     unset ANTHROPIC_API_KEY CLAUDE_CODE_OAUTH_TOKEN
-    # Pin every model slot to a grok coding model. GROK_MODEL overrides — e.g. set it
-    # to grok-4.5 (xAI's current flagship, available on api.x.ai) instead of the
-    # conservative grok-build-0.1 default kept here for backward-compatible behavior.
-    grok_model="${GROK_MODEL:-grok-build-0.1}"
+    # Pin every model slot to a grok coding model. Defaults to grok-4.5, xAI's current
+    # flagship and the same id the grok harness offers (GROK_MODELS in the dashboard's
+    # constants.ts), so the gateway and CLI paths name the same model. GROK_MODEL
+    # overrides: the older coding ids (grok-build-0.1, grok-composer-2.5-fast,
+    # grok-4.3) are api.x.ai model strings that still work here, even though the grok
+    # CLI rejects them on an X-account OAuth login.
+    grok_model="${GROK_MODEL:-grok-4.5}"
     export ANTHROPIC_DEFAULT_OPUS_MODEL="$grok_model"
     export ANTHROPIC_DEFAULT_SONNET_MODEL="$grok_model"
     export ANTHROPIC_DEFAULT_HAIKU_MODEL="$grok_model"
@@ -236,31 +239,33 @@ case "${GATEWAY:-direct}" in
     # uses dot-form ids: drop any trailing -YYYYMMDD date, then convert each
     # <digit>-<digit> to <digit>.<digit>. SURPLUS_MODEL overrides; opus-4.8 is the
     # fallback when $MODEL is unset.
-    surplus_model="${SURPLUS_MODEL:-$(printf '%s' "${MODEL:-claude-opus-4-8}" | sed -E 's/-[0-9]{8}$//; s/([0-9])-([0-9])/\1.\2/g')}"
+    surplus_model="${SURPLUS_MODEL:-$(printf '%s' "${MODEL:-claude-opus-5}" | sed -E 's/-[0-9]{8}$//; s/([0-9])-([0-9])/\1.\2/g')}"
     start_ccr_sidecar surplus \
       "https://www.surplusintelligence.ai/api/inference/v1/chat/completions" \
       "$SURPLUS_API_KEY" "$surplus_model"
     echo "::notice::Routing through Surplus via claude-code-router (${surplus_model})"
     ;;
 
-  venice)  # SIDECAR — OpenAI-compatible (dash-form ids); tops out ~Opus 4.6
+  venice)  # SIDECAR — OpenAI-compatible (dash-form ids); carries Opus 5, no haiku
     require_secret VENICE_API_KEY
     # Set VENICE_CLEANCACHE=1 to add the cleancache transformer (1h TTL, avoids
     # the shared 4-block prompt-cache limit) if you hit cache errors.
-    # The sidecar pins ONE model, so track aeon's $MODEL — but Venice's catalog
-    # caps at ~Opus 4.6, so a newer $MODEL (opus 4.7/4.8, fable) would 404. Derive
-    # only for models Venice is known to carry (dash-form, date suffix stripped);
-    # anything else keeps the safe opus-4-6 default. VENICE_MODEL overrides.
-    # (Confirm/extend the allowlist when Venice is live-validated.)
+    # The sidecar pins ONE model, so track aeon's $MODEL. Venice names models with
+    # aeon's own dash-form ids, so the picker's ids pass straight through (date
+    # suffix stripped) when Venice carries them. It carries NO haiku at all, so
+    # haiku, and anything else off-catalog, falls back to sonnet-5 rather than
+    # 404ing on a model Venice never had. VENICE_MODEL overrides.
+    # Allowlist verified against api.venice.ai/api/v1/models on 2026-07-24.
     # VENICE_BASE_URL (repo variable) points the sidecar at any Venice-compatible
     # endpoint — a self-hosted relay, a billing proxy, a regional mirror — same
     # override pattern as VENICE_MODEL. Defaults to Venice's public API.
     venice_model="${VENICE_MODEL:-}"
     if [ -z "$venice_model" ]; then
       m="$(printf '%s' "${MODEL:-}" | sed -E 's/-[0-9]{8}$//')"
       case "$m" in
-        claude-opus-4-6|claude-sonnet-4-6|claude-haiku-4-5) venice_model="$m" ;;
-        *) venice_model="claude-opus-4-6" ;;
+        claude-opus-5|claude-sonnet-5|claude-fable-5|claude-opus-4-8|claude-opus-4-7|\
+        claude-opus-4-6|claude-opus-4-5|claude-sonnet-4-6|claude-sonnet-4-5) venice_model="$m" ;;
+        *) venice_model="claude-sonnet-5" ;;
       esac
     fi
     start_ccr_sidecar venice \

scripts/tests/test_run_grok.sh

@@ -86,7 +86,7 @@ grep -qx -- "--permission-mode" "$ARGS_FILE" && grep -qx "bypassPermissions" "$A
 grep -qx -- "--rules" "$ARGS_FILE" && pass "passes --rules compat preamble" || bad "passes --rules compat preamble"
 
 # 4b. --model is OMITTED for a leftover claude-* id or empty (grok uses its default)
-echo "p" | GROK_FAKE_OUT='{"text":"x"}' MODEL=claude-sonnet-4-6 SKILL_MODE=write XAI_API_KEY=xai-test bash "$R" >/dev/null 2>&1
+echo "p" | GROK_FAKE_OUT='{"text":"x"}' MODEL=claude-sonnet-5 SKILL_MODE=write XAI_API_KEY=xai-test bash "$R" >/dev/null 2>&1
 if grep -qx -- "--model" "$ARGS_FILE"; then bad "omits --model for claude-* id"; else pass "omits --model for claude-* id"; fi
 echo "p" | GROK_FAKE_OUT='{"text":"x"}' MODEL="" SKILL_MODE=write XAI_API_KEY=xai-test bash "$R" >/dev/null 2>&1
 if grep -qx -- "--model" "$ARGS_FILE"; then bad "omits --model for empty model"; else pass "omits --model for empty model"; fi