diff --git a/README.md b/README.md
index 7c01566..deb4f2b 100644
--- a/README.md
+++ b/README.md
@@ -221,7 +221,7 @@ For the common retry flow with last-turn steering or stable default timeouts, op
 ```ts
 const review = agent({
   maxTurns: 3,
-  addons: [timeout({ timeout: 30_000 }), steering(), repair],
+  addons: [timeout({ timeout: 30_000 }), steering(), repair()],
 });
 ```
 
diff --git a/docs/rig-syntax-copilot-pi-agent-comparison.md b/docs/rig-syntax-copilot-pi-agent-comparison.md
index 5df8c5e..25139f6 100644
--- a/docs/rig-syntax-copilot-pi-agent-comparison.md
+++ b/docs/rig-syntax-copilot-pi-agent-comparison.md
@@ -15,7 +15,7 @@ The focus is generation reliability: what an agent can produce quickly with low
 | `s.object(...)`, `s.enum(...)`, `s.array(...)` | Explicit JSON schema or prompt-constrained JSON validated in app code | Same pattern: schema-constrained JSON validated by the harness/app |
 | `p.read(...)`, `p.bash(...)` | Tool/context calls orchestrated by the host app before/within turns | Tool/context calls via pi-agent tool integration/orchestration |
 | `agents: { subagent }` | Multiple sessions/roles coordinated in app orchestration | Multi-agent graph/delegation orchestration |
-| `maxTurns`, optional `rig/addons` repair addon | Explicit retry + repair loop in app logic | Retry/repair policies in agent workflow/harness |
+| `maxTurns`, optional `repair()` addon | Explicit retry + repair loop in app logic | Retry/repair policies in agent workflow/harness |
 | `permissions` | Host-side policy gates around shell/write operations | Host-side tool permission policies |
 
 ## 2) Top 10 scenarios: Copilot SDK APIs (ranked easiest → hardest)
@@ -30,7 +30,7 @@ The focus is generation reliability: what an agent can produce quickly with low
 | 6 | PR triage recommendation | Requires prioritization judgment and policy interpretation. | One/two turns with constrained triage schema and confidence fields. |
 | 7 | README draft generation | Creative synthesis adds style and completeness ambiguity. | Multi-section structured output with post-parse checks. |
 | 8 | Release notes generation | Requires grouping/dedup across many commits. | Batched commit input + grouped typed output contract. |
-| 9 | Schema-repairing extractor | Needs robust retry when output is invalid or partial. | `maxTurns` plus optional `rig/addons` repair addon maps to explicit app-level validation/repair loop. |
+| 9 | Schema-repairing extractor | Needs robust retry when output is invalid or partial. | `maxTurns` plus optional `repair()` addon maps to an explicit app-level validation/repair loop. |
 | 10 | Multi-agent orchestrator | Highest coordination overhead across roles and merges. | `agents` maps to multi-session orchestration and aggregation logic. |
 
 ## 3) Top 10 scenarios: pi-agent SDK (ranked easiest → hardest)
diff --git a/skills/rig/SKILL.md b/skills/rig/SKILL.md
index 78e3482..7728c1d 100644
--- a/skills/rig/SKILL.md
+++ b/skills/rig/SKILL.md
@@ -93,11 +93,11 @@ Do not replace file intents with `cat` commands or large in-memory strings. `p.w
 
 ## Tools, composition, and reliability
 
-- Define tools with `defineTool(name, { description, parameters: s.object(...), handler })`; schema-based handler arguments are inferred and tools default to `skipPermission: true`.
+- Define tools with `defineTool(name, { description, parameters: s.object(...), handler })`; schema-based handler arguments are inferred and tools default to `skipPermission: true`. Destructure only handler fields you use.
 - `agents` is a named object such as `agents: { extractor }`, never an array. Attach every declared subagent to the exported root's graph.
 - There is no chain or loop primitive; give the coordinator explicit delegation instructions and require one combined output.
-- Automatic parse/schema repair requires `repair` from `rig/addons`; `maxTurns` alone only sets the total turn budget.
-- Put `steering()` after `repair` when adding a final-turn warning. Use `oncePerAgent()` for one registration callback per runtime agent.
+- Automatic parse/schema repair requires `repair()` from `rig/addons`; put its `maxTurns` budget on the agent spec.
+- For a final-turn warning, use `addons: [steering(), repair()]`. Custom text uses `steering({ message: "..." })`, not a positional string. Use `oncePerAgent()` for one registration callback per runtime agent.
 
 ## Runnable markdown
 
diff --git a/skills/rig/addons.ts b/skills/rig/addons.ts
index 76e6bf7..3dc1949 100644
--- a/skills/rig/addons.ts
+++ b/skills/rig/addons.ts
@@ -4,6 +4,7 @@ import type { Agent, AgentAddon, AgentAddonContext } from "./rig.ts";
 const DEFAULT_STEERING_WARNING = "You are running out of turns. This is your final attempt before reaching the turn limit. Please correct your output now.";
 
 export type SteeringOptions = {
+  /** Warning appended to the final retry prompt. */
   message?: string;
 };
 
@@ -16,6 +17,12 @@ export type AgentRegistration = (
   context: AgentAddonContext,
 ) => void | Promise<void>;
 
+/**
+ * Appends a final-attempt warning to the retry prompt produced by an inner addon.
+ *
+ * Place this before `repair()`, for example
+ * `addons: [steering({ message: "Return valid JSON now." }), repair()]`.
+ */
 export function steering(options: SteeringOptions = {}): AgentAddon {
   const message = options.message ?? DEFAULT_STEERING_WARNING;
   return async (context, next) => {
@@ -26,26 +33,33 @@ export function steering(options: SteeringOptions = {}): AgentAddon {
   };
 }
 
-export const repair: AgentAddon = async (context, next) => {
-  await next();
-  if (context.completed || context.error !== undefined || context.nextPrompt !== undefined) {
-    return;
-  }
-  if (context.response === undefined) {
-    return;
-  }
-  const analysis = analyzeResponse(context.response, context.outputSchema, context.spec.name, context.turn);
-  if (analysis.ok) {
-    context.completed = true;
-    context.output = analysis.output;
-    return;
-  }
-  if (context.turn >= context.maxTurns) {
-    context.error = analysis.error;
-    return;
-  }
-  context.nextPrompt = defaultRepairPrompt(context.spec, analysis.error);
-};
+/**
+ * Parses and validates responses, retrying failures within the agent's turn budget.
+ *
+ * Configure `maxTurns` on the agent spec.
+ */
+export function repair(): AgentAddon {
+  return async (context, next) => {
+    await next();
+    if (context.completed || context.error !== undefined || context.nextPrompt !== undefined) {
+      return;
+    }
+    if (context.response === undefined) {
+      return;
+    }
+    const analysis = analyzeResponse(context.response, context.outputSchema, context.spec.name, context.turn);
+    if (analysis.ok) {
+      context.completed = true;
+      context.output = analysis.output;
+      return;
+    }
+    if (context.turn >= context.maxTurns) {
+      context.error = analysis.error;
+      return;
+    }
+    context.nextPrompt = defaultRepairPrompt(context.spec, analysis.error);
+  };
+}
 
 export function timeout(options: TimeoutOptions): AgentAddon {
   return async (context, next) => {
diff --git a/skills/rig/references/agent-api.md b/skills/rig/references/agent-api.md
index 9bb1e66..665fe65 100644
--- a/skills/rig/references/agent-api.md
+++ b/skills/rig/references/agent-api.md
@@ -137,6 +137,8 @@ export default triage;
 
 For plain JSON Schema parameters, provide a generic such as `defineTool<{ issue: string }>(...)`. A handler may return a string or any JSON-serializable value; Rig serializes non-string values, so do not call `JSON.stringify` in the handler. Tools default to `skipPermission: true`.
 
+Strict TypeScript compilation reports unused handler bindings. Destructure only the keys the handler uses, or rename an unavoidable binding with a leading underscore, such as `{ filename: _filename, content }`.
+
 ## Call-time options
 
 Use call-time options only for per-run changes:
@@ -161,6 +163,6 @@ Use only the current API:
 - `agent({ name, ... })`
 - `p.*` and ``p`...` `` from `rig`
 - `s.*` for explicit schemas
-- `oncePerAgent`, `repair`, `steering`, and `timeout` from `rig/addons`
+- `oncePerAgent`, `repair()`, `steering`, and `timeout` from `rig/addons`
 
 Do not add deprecated hooks, alternate schema syntaxes, or compatibility bridges.
diff --git a/skills/rig/references/composition.md b/skills/rig/references/composition.md
index 7b45a8f..c560de7 100644
--- a/skills/rig/references/composition.md
+++ b/skills/rig/references/composition.md
@@ -100,7 +100,7 @@ When a task asks for runnable markdown:
 
 ## Repair
 
-Rig starts with no addons. `maxTurns` is only the total budget; automatic parse/schema correction requires `repair`:
+Rig starts with no addons. `maxTurns` is only the total budget; automatic parse/schema correction requires `repair()`:
 
 ```ts
 import { agent } from "rig";
@@ -110,17 +110,17 @@ import { repair } from "rig/addons";
 const summarize = agent({
   model: "mini",
   maxTurns: 3,
-  addons: repair,
+  addons: repair(),
 });
 
 export default summarize;
 ```
 
-The budget includes the initial attempt and all retries.
+The budget includes the initial attempt and all retries. Configure `maxTurns` on the agent spec; a call-time value can override it.
 
 ## Final-turn steering
 
-`steering()` appends a last-chance warning to the final retry prompt produced by `repair`. Put it after `repair`:
+`steering()` appends a last-chance warning to the final retry prompt produced by `repair`. Put it before `repair` so it can observe the repair prompt as the addon chain unwinds:
 
 ```ts
 import { agent } from "rig";
@@ -130,13 +130,13 @@ import { repair, steering } from "rig/addons";
 const summarize = agent({
   model: "mini",
   maxTurns: 3,
-  addons: [repair, steering()],
+  addons: [steering(), repair()],
 });
 
 export default summarize;
 ```
 
-Use `repair` alone when the validation error is enough guidance. Do not use `steering()` without `repair`, because it only augments prompts generated by repair.
+Use `repair()` alone when the validation error is enough guidance. Pass custom warning text in an options object, as in `steering({ message: "Return valid JSON now." })`; a positional string is invalid. Do not use `steering()` without `repair()`, because it only augments prompts generated by repair.
 
 ## One-time runtime registration
 
@@ -155,7 +155,7 @@ const qa = agent({
     oncePerAgent((runtimeAgent) => {
       initializedAgents.add(runtimeAgent);
     }),
-    repair,
+    repair(),
   ],
 });
 
diff --git a/skills/rig/rig.ts b/skills/rig/rig.ts
index 2db72e2..f3f9701 100644
--- a/skills/rig/rig.ts
+++ b/skills/rig/rig.ts
@@ -595,7 +595,7 @@ export type AgentSpec<Input extends Schema = StringSchema, Output extends Schema
   model?: string;
   /** Maximum number of turns (initial + repair retries). Defaults to `4`. */
   maxTurns?: number;
-  /** Middleware addons that wrap each turn's ask/response cycle, e.g. `repair`, `steering`. */
+  /** Middleware addons that wrap each turn's ask/response cycle, e.g. `repair()`, `steering()`. */
   addons?: AgentAddon | AgentAddon[];
   /** Named sub-agents available for delegation from this agent's prompt. */
   agents?: Record<string, AgentFn<any, any>>;
diff --git a/skills/rig/samples/66-ci-workflow-health.md b/skills/rig/samples/66-ci-workflow-health.md
index 0a7a402..84212ec 100644
--- a/skills/rig/samples/66-ci-workflow-health.md
+++ b/skills/rig/samples/66-ci-workflow-health.md
@@ -21,7 +21,7 @@ const parseYamlSteps = defineTool("parse_yaml_steps", {
 const ciWorkflowAnalyzer = agent({
   model: "mini",
   maxTurns: 3,
-  addons: repair,
+  addons: repair(),
   instructions: p`Scan ${p.glob(".github/workflows/*.yml")} and ${p.bashRaw`find .github/workflows -name '*.yaml' 2>/dev/null`} for workflow health issues. Use parse_yaml_steps to inspect individual files.`,
   output: s.object({
     issues: s.array(s.object({ workflow: s.nonEmptyString, severity: s.enum("info", "warning", "error"), message: s.string, fix: s.optional(s.string) })),
@@ -32,4 +32,3 @@ const ciWorkflowAnalyzer = agent({
 
 export default ciWorkflowAnalyzer;
 ```
-
diff --git a/src/rig.test.ts b/src/rig.test.ts
index c3fd09e..7dc7f8d 100644
--- a/src/rig.test.ts
+++ b/src/rig.test.ts
@@ -399,7 +399,7 @@ describe("agent invocation", () => {
 
     const repairable = agent({
       name: "repairable",
-      addons: repair,
+      addons: repair(),
       maxTurns: 2,
     });
 
@@ -458,7 +458,7 @@ describe("agent invocation", () => {
             context.nextPrompt = `please fix: ${context.nextPrompt}`;
           }
         },
-        repair,
+        repair(),
       ],
       maxTurns: 2,
     });
@@ -700,7 +700,7 @@ describe("agent invocation", () => {
             context.nextPrompt = `${context.nextPrompt}\nAdd a short correction because you are running out of turns.`;
           }
         },
-        repair,
+        repair(),
       ],
     });
 
@@ -727,7 +727,7 @@ describe("agent invocation", () => {
     const steerable = agent({
       name: "steerable",
       maxTurns: 2,
-      addons: [steering(), repair],
+      addons: [steering(), repair()],
     });
 
     await expect(steerable("go")).resolves.toBe("recovered");
@@ -760,7 +760,7 @@ describe("agent invocation", () => {
             }
           }
         },
-        repair,
+        repair(),
       ],
     });
 
@@ -790,7 +790,7 @@ describe("agent invocation", () => {
         oncePerAgent(async (runtimeAgent, context) => {
           register(runtimeAgent, context.turn);
         }),
-        repair,
+        repair(),
       ],
     });