@massa-ai/cursor-plugin 1.60.0 → 1.61.0

package/.cursor-plugin/plugin.json

@@ -1,5 +1,5 @@
 {
   "name": "massa-ai",
-  "version": "1.60.0",
+  "version": "1.61.0",
   "description": "massa-ai — semantic code search, memory, and context compression for Cursor"
 }

package/agent-profiles/balanced/massa-ai-judge.md

@@ -77,9 +77,9 @@ next_step: <string>
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot
-assignments.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot assignments.
 
 ## Validation Sensors
 - Every criterion score carries an exact quotation from the artifact.

package/agent-profiles/balanced/massa-ai-meta-judge.md

@@ -69,8 +69,9 @@ overall: weighted-mean
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `meta-judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
 
 ## Validation Sensors
 - Output parses as YAML; weights sum to 1.0 (±0.001); every criterion carries id, name, weight, scale (min 1, max 5), rubric anchors for 1/3/5, and a non-empty checklist.

package/agent-profiles/cheap/massa-ai-judge.md

@@ -77,9 +77,9 @@ next_step: <string>
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot
-assignments.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot assignments.
 
 ## Validation Sensors
 - Every criterion score carries an exact quotation from the artifact.

package/agent-profiles/cheap/massa-ai-meta-judge.md

@@ -69,8 +69,9 @@ overall: weighted-mean
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `meta-judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
 
 ## Validation Sensors
 - Output parses as YAML; weights sum to 1.0 (±0.001); every criterion carries id, name, weight, scale (min 1, max 5), rubric anchors for 1/3/5, and a non-empty checklist.

package/agent-profiles/heavy/massa-ai-judge.md

@@ -77,9 +77,9 @@ next_step: <string>
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot
-assignments.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot assignments.
 
 ## Validation Sensors
 - Every criterion score carries an exact quotation from the artifact.

package/agent-profiles/heavy/massa-ai-meta-judge.md

@@ -69,8 +69,9 @@ overall: weighted-mean
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `meta-judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
 
 ## Validation Sensors
 - Output parses as YAML; weights sum to 1.0 (±0.001); every criterion carries id, name, weight, scale (min 1, max 5), rubric anchors for 1/3/5, and a non-empty checklist.

package/agent-profiles/home/massa-ai-judge.md

@@ -77,9 +77,9 @@ next_step: <string>
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot
-assignments.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot assignments.
 
 ## Validation Sensors
 - Every criterion score carries an exact quotation from the artifact.

package/agent-profiles/home/massa-ai-meta-judge.md

@@ -69,8 +69,9 @@ overall: weighted-mean
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `meta-judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
 
 ## Validation Sensors
 - Output parses as YAML; weights sum to 1.0 (±0.001); every criterion carries id, name, weight, scale (min 1, max 5), rubric anchors for 1/3/5, and a non-empty checklist.

package/agent-profiles/work/massa-ai-judge.md

@@ -77,9 +77,9 @@ next_step: <string>
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot
-assignments.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot assignments.
 
 ## Validation Sensors
 - Every criterion score carries an exact quotation from the artifact.

package/agent-profiles/work/massa-ai-meta-judge.md

@@ -69,8 +69,9 @@ overall: weighted-mean
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `meta-judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
 
 ## Validation Sensors
 - Output parses as YAML; weights sum to 1.0 (±0.001); every criterion carries id, name, weight, scale (min 1, max 5), rubric anchors for 1/3/5, and a non-empty checklist.

package/agents/massa-ai-judge.md

@@ -77,9 +77,9 @@ next_step: <string>
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot
-assignments.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot assignments.
 
 ## Validation Sensors
 - Every criterion score carries an exact quotation from the artifact.

package/agents/massa-ai-meta-judge.md

@@ -69,8 +69,9 @@ overall: weighted-mean
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `meta-judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
 
 ## Validation Sensors
 - Output parses as YAML; weights sum to 1.0 (±0.001); every criterion carries id, name, weight, scale (min 1, max 5), rubric anchors for 1/3/5, and a non-empty checklist.

package/hooks/massa-ai-hook

@@ -187,6 +187,93 @@ export function resolveHookApiKey(): string {
   return "";
 }
 
+// ── Session-start drift line (agent-runtime-drift T06) ─────────────────────
+
+/** Same XDG resolution as getHookConfigPath — duplicated for the same
+ *  dependency-freeze reason, pinned by tests on both sides. */
+export function getInstallStatePath(): string {
+  const xdg = process.env.XDG_CONFIG_HOME;
+  const base = xdg && xdg.trim() ? xdg : path.join(homedir(), ".config");
+  return path.join(base, "massa-ai", "install-state.json");
+}
+
+function readJsonAt(filePath: string): Record<string, unknown> | null {
+  try {
+    const parsed = JSON.parse(readFileSync(filePath, "utf8"));
+    return parsed && typeof parsed === "object" && !Array.isArray(parsed)
+      ? (parsed as Record<string, unknown>)
+      : null;
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Pinned one-key read (`^model:`) — deliberately NOT the shared frontmatter
+ * parser: this binary's dependency freeze (see getHookConfigPath) forbids
+ * importing @massa-ai/shared. The cross-side test pins this regex against the
+ * shared parser's output (spec INV B3), so the writer and this reader cannot
+ * silently disagree about what `model:` means.
+ */
+function readModelLine(filePath: string): string | null {
+  try {
+    const match = /^model:\s*(.+)\r?$/m.exec(readFileSync(filePath, "utf8"));
+    const value = match?.[1]?.trim();
+    return value ? value.replace(/^["']|["']$/g, "") : null;
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * The ≤2-line session-start drift report (agent-runtime-drift): version
+ * drift (live bundle vs recorded state), agent drift (active investigator
+ * model vs the recorded profile's variant), and the host env override that
+ * nullifies every per-agent model at runtime. Null when every recording
+ * agrees — silent when healthy. Read-only (INV5); every failure degrades to
+ * fewer lines, never a throw. The fix path is the existing profile switch,
+ * never the hook.
+ */
+export function buildSessionStartDoctorLine(
+  pluginRoot: string,
+  env: Readonly<Record<string, string | undefined>>,
+): string | null {
+  const lines: string[] = [];
+
+  const sourceVersion = readJsonAt(path.join(pluginRoot, ".claude-plugin", "plugin.json"))?.version;
+  const state = readJsonAt(getInstallStatePath());
+  const platforms = state?.platforms as Record<string, Record<string, unknown>> | undefined;
+  const claude = platforms?.claude;
+  const stateVersion = claude?.plugin?.version;
+  if (typeof sourceVersion === "string" && typeof stateVersion === "string" && sourceVersion !== stateVersion) {
+    lines.push(
+      `[massa-ai] version drift: live bundle ${sourceVersion} vs recorded ${stateVersion} — update the plugin or re-run the installer`,
+    );
+  }
+
+  const profile = claude?.modelProfile?.profile;
+  if (typeof profile === "string" && profile) {
+    const activeModel = readModelLine(path.join(pluginRoot, "agents", "massa-ai-investigator.md"));
+    const variantModel = readModelLine(
+      path.join(pluginRoot, "agent-profiles", profile, "massa-ai-investigator.md"),
+    );
+    if (activeModel && variantModel && activeModel !== variantModel) {
+      lines.push(
+        `[massa-ai] agent drift: investigator ${activeModel} vs ${profile} variant ${variantModel} — re-run the profile switch`,
+      );
+    }
+  }
+
+  const override = env.CLAUDE_CODE_SUBAGENT_MODEL;
+  if (typeof override === "string" && override.trim()) {
+    lines.push(
+      `[massa-ai] CLAUDE_CODE_SUBAGENT_MODEL=${override.trim()} overrides every per-agent model at runtime`,
+    );
+  }
+
+  return lines.length ? lines.slice(0, 2).join("\n") : null;
+}
+
 // ── POST helper ─────────────────────────────────────────────────────────────
 
 export function postObservation(
@@ -299,6 +386,17 @@ export async function main(stdinInput?: string): Promise<void> {
 
   const projectId = resolveProjectId(sessionId, cwd);
 
+  // agent-runtime-drift (T06): session-start prints a bounded drift line when
+  // the recordings disagree (read-only, silent when healthy — INV5).
+  if (subcommand === "session-start") {
+    try {
+      const doctorLine = buildSessionStartDoctorLine(path.resolve(import.meta.dirname, ".."), process.env);
+      if (doctorLine) process.stdout.write(doctorLine + "\n");
+    } catch {
+      // silent-degrade: drift reporting must never block the agent
+    }
+  }
+
   const baseUrl = process.env.MASSA_AI_API_BASE || "http://localhost:3333";
   const hookUrl = `${baseUrl}/api/v1/hook`;
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@massa-ai/cursor-plugin",
-  "version": "1.60.0",
+  "version": "1.61.0",
   "description": "massa-ai plugin for Cursor — semantic code search, memory, and context compression",
   "files": [
     "agents",

package/skills/agents/architecture-specialist/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/audit-specialist/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/builder/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: standard
   permission: write
 ---
 

package/skills/agents/context-curator/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/designer/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: standard
   permission: write
 ---
 

package/skills/agents/documentation-agent/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: light
   permission: write
 ---
 

package/skills/agents/furps-analyst/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/investigator/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/judge/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.2.0"
-  model_tier: deep
   permission: write
 ---
 
@@ -83,9 +82,9 @@ next_step: <string>
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot
-assignments.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the per-slot fallback; `workflows/judge-with-debate.md` owns the live slot assignments.
 
 ## Validation Sensors
 - Every criterion score carries an exact quotation from the artifact.

package/skills/agents/meta-judge/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.1.0"
-  model_tier: deep
   permission: read-only
 ---
 
@@ -74,8 +73,9 @@ overall: weighted-mean
 - References (paths relative to the `massa-ai` skill directory): `references/agent-orchestration.md`, `references/audit-report-io.md` (Judge With Debate Report Contracts).
 
 ## Model Hint
-See `references/agent-orchestration.md` (Model Diversity Fallback): `metadata.model_tier`
-(`deep`) is the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
+See `references/agent-orchestration.md` (Model Diversity Fallback): `meta-judge` carries no
+per-agent override in `skills/model-profiles.json`, so the active profile's host default is
+the fallback; `workflows/judge-with-debate.md` owns the live model assignment.
 
 ## Validation Sensors
 - Output parses as YAML; weights sum to 1.0 (±0.001); every criterion carries id, name, weight, scale (min 1, max 5), rubric anchors for 1/3/5, and a non-empty checklist.

package/skills/agents/mobile-specialist/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/navigator/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.1.0"
-  model_tier: deep
   permission: read-only
   tools: mcp-index
 ---

package/skills/agents/plan-critic/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/planner/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/requirements-analyst/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/reviewer/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/agents/test-engineer/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: standard
   permission: write
 ---
 

package/skills/agents/verification-agent/SKILL.md

@@ -5,7 +5,6 @@ license: MIT
 metadata:
   author: Luiz Massa
   version: "1.0.0"
-  model_tier: deep
   permission: read-only
 ---
 

package/skills/massa-ai/references/agent-orchestration.md

@@ -117,18 +117,17 @@ delegation with its reason in the Evidence Gate.
 
 ## Model Diversity Fallback
 
-Applies to any charter whose `metadata.model_tier` is a fallback and whose
-dispatching workflow additionally requests per-invocation model diversity at
-dispatch time (e.g. `judge` — 3 parallel slots; `meta-judge` — one slot).
-
-- The charter's `metadata.model_tier` is the fallback every host runs when
-  dispatch-time model selection is unavailable.
-- The dispatching workflow (e.g. `workflows/judge-with-debate.md`) is the
-  single source for the current slot/model assignment, not the charter file.
-- When dispatch-time selection is unavailable, every affected slot runs the
-  charter default, and the orchestrator records `DIVERSITY DEGRADED`
-  (multi-slot) or an equivalent diversity warning (single-slot) per the
-  dispatching workflow's own contract.
+Applies to any charter whose dispatching workflow requests per-invocation model diversity
+at dispatch time (e.g. `judge` — 3 parallel slots; `meta-judge` — one slot).
+
+- Read-only agents (judge, meta-judge, audit-specialist, navigator, and others) carry no
+  per-agent model override in the profile and resolve to the profile's per-tool default.
+  By convention, the default is the profile's strongest model.
+- The dispatching workflow (e.g. `workflows/judge-with-debate.md`) is the single source
+  for the current slot/model assignment, not the charter file.
+- When dispatch-time selection is unavailable, every affected slot runs the profile default,
+  and the orchestrator records `DIVERSITY DEGRADED` (multi-slot) or an equivalent diversity
+  warning (single-slot) per the dispatching workflow's own contract.
 
 ## Roles
 
@@ -303,7 +302,7 @@ like every other dispatch.
 ```md
 🤖 [Agent Started] Investigator — model opus, effort high. Scope: the four emitters.
 🤖 [Agent Started] Designer — model/effort unknown (no installed agent file at
-   ~/.claude/plugins/cache/massa-ai/massa-ai/1.48.0/agents/massa-ai-designer.md). Dispatching anyway.
+   <liveRoot>/agents/massa-ai-designer.md). Dispatching anyway.
 ```
 
 That second line is a measured case, not a hypothetical: on a machine with plugin
@@ -316,7 +315,7 @@ this table so the two cannot drift silently:
 
 | Host | Installed agents directory | Glob | Model / effort keys |
 | --- | --- | --- | --- |
-| Claude — marketplace route | `<marketplaceRoot>/agents` — a *versioned* bundle root, e.g. `~/.claude/plugins/cache/massa-ai/massa-ai/1.48.0/agents` | `massa-ai-*.md` | `model:` / `effort:` |
+| Claude — marketplace route | `<marketplaceRoot>/agents`, where `<marketplaceRoot>` is `resolveClaudeMarketplaceInstall`'s live root: for a **directory-source** marketplace the host loads the plugin LIVE from the source bundle — e.g. `<repo>/apps/claude-plugin/agents`; for any other kind it is the *versioned* cache snapshot, e.g. `~/.claude/plugins/cache/massa-ai/massa-ai/1.48.0/agents` (a stale-able snapshot — never hardcode it; read `profile_list`'s `liveRoot`) | `massa-ai-*.md` | `model:` / `effort:` |
 | Claude — file route | `~/.claude/agents` | `massa-ai-*.md` | `model:` / `effort:` |
 | Codex | `~/.codex/agents` | `massa-ai-*.toml` | `model` / `model_reasoning_effort` |
 | OpenCode | `~/.config/opencode/agents` | `massa-ai-*.md` | `model:` / `reasoningEffort:` |

package/skills/massa-ai/references/spec-driven/sub-agents.md

@@ -178,7 +178,7 @@ When sub-agents are unavailable (a single agent executing the full feature), use
 
 **Applies only if the harness can assign a model per sub-agent.** If it cannot, ignore this section and run everything on the default model — the workflow is correct either way. The point is to spend high-reasoning capacity where ambiguity and consequence are high, and a faster tier where the work is mechanical, instead of paying top-tier cost uniformly.
 
-massa-ai resolves the actual model per agent through `metadata.model_tier` (`light` / `standard` / `deep`) in each sub-agent's charter (`skills/agents/<name>/SKILL.md`), combined with the host and the active profile in `skills/model-profiles.json` (see `CLAUDE.md` § Agent-harness surface). This section maps role/work characteristics onto that mechanism — it is not a separate free-floating table.
+massa-ai resolves the actual model per agent through the active profile in `skills/model-profiles.json` (see `CLAUDE.md` § Agent-harness surface). Most agents use the profile's per-tool default; some carry per-agent overrides. This section maps role/work characteristics onto that decision framework — it is not a separate free-floating table.
 
 Judge the tier by the work in front of the role, not by the role's title:
 
@@ -195,7 +195,7 @@ Judge the tier by the work in front of the role, not by the role's title:
 **Rules of thumb:**
 
 - When unsure, size up, not down. An under-powered worker on ambiguous logic produces gaps the Verifier then has to catch — more expensive than paying for reasoning once.
-- **The Verifier always runs on the deepest tier** — per project rule, `skills/agents/verification-agent/SKILL.md` pins `metadata.model_tier: deep`, structurally, not just as advisory guidance here. A weak Verifier defeats the author ≠ verifier gate.
-- **Read-only specialists always run on the deepest tier** — this generalizes the Verifier rule: every findings-only or investigation-only charter (`permission: read-only`) pins `metadata.model_tier: deep` structurally, because there is no later implementation pass to catch what a weaker read-only pass missed.
+- **The Verifier always runs on the deepest tier** — in built-in profiles, the verification-agent carries no per-agent override and resolves to the profile's strongest model. A weak Verifier defeats the author ≠ verifier gate.
+- **Read-only specialists always run on the deepest tier** — this generalizes the Verifier rule: in built-in profiles, every findings-only or investigation-only charter (`permission: read-only`) carries no per-agent override and resolves to the profile's strongest model, because there is no later implementation pass to catch what a weaker read-only pass missed. **Accepted risk:** user overlays via the Web UI Model Catalog can choose a weaker default; the UI help text documents this convention.
 - Set the tier per batch, from that batch's phases. A feature can mix tiers across batches.
 - Outside the Verifier's and read-only specialists' structural pins, this table is advisory metadata only — no gate, commit, or verification step depends on it.

package/skills/massa-ai/references/synapse-policy.md

@@ -44,7 +44,7 @@ Config knobs (env, read by the server, not by the agent):
 |---|---|---|
 | `SYNAPSE_ENABLED` | `true` | Master kill switch; `false` bypasses the whole pipeline, and `search` behaves statelessly even when a `sessionId` is passed. |
 | `SYNAPSE_ATTENTION_ENABLED` | `false` | Multi-signal attention re-ranker. **Off by default** — do not attribute re-ranking to Synapse unless it is on. |
-| `LOG_LEVEL` | `info` | `debug` emits one pipeline log line per query (see Reading Pipeline Output). |
+| `MASSA_AI_LOG_LEVEL` | `info` | `debug` emits one pipeline log line per query (see Reading Pipeline Output). |
 
 ## MCP-First Lifecycle
 
@@ -103,7 +103,7 @@ optional. REST prefetch requires `filePath` and may include `symbols`, `chains`,
 
 ## Reading Pipeline Output
 
-With `LOG_LEVEL=debug`, the server emits one structured line per Synapse-scoped
+With `MASSA_AI_LOG_LEVEL=debug`, the server emits one structured line per Synapse-scoped
 query. Use it to decide whether Synapse is helping or whether the query needs
 refining — not as evidence about the codebase.