diff --git a/CHANGELOG.md b/CHANGELOG.md
index 90c0ea9f0..321c0d7a5 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,9 +1,21 @@
 # Changelog
 
-## 1.0.2-beta.0 — 2026-08-19
+## 1.0.2-beta.0 — 2026-08-20
+
+### Docs
+
+- Reposition the project from "runtime failure resolution for coding agents" to observability and enforcement for agent harnesses, across the surfaces that render its identity. The product was never coding-only — `INTEGRATION_TYPES` holds twelve harnesses and two of them (Hermes, a Slack/Telegram gateway; OpenClaw, a self-hosted assistant) were never coding tools — but the strings a stranger reads said otherwise, and the breadth evidence sat under labels that converted it back: README's twelve-logo grid, containing both gateways, was headed "Supported agent CLIs". The README hero, `package.json`'s description and keywords, `SECURITY.md`, `types.ts`, the docs index and quickstart, `reference/harnesses`, the sessions captions, the local dashboard, and the audit share templates now name the harness and define it ostensively on first use, with coding CLIs framed as the class covered deepest rather than the category. `reference/harnesses` is promoted onto the onboarding path in all fifteen locale navs, since it is the only page that answers which kinds of agent are covered and it sat in the last tab. (#731)
+
+- Correct three counts that had rotted silently, and add the test that stops it recurring. README advertised "30 built-in policies" in three places; the real number is 40, and had been for some time. `__tests__/scripts/copy-counts.test.ts` now derives every count in copy — harnesses, policies, canonical events, and the harnesses on which `PreToolUse` and `Stop` blocking is verified — from `types.ts`, `builtin-policies.ts` and `enforcement-capability.ts`, and fails when prose disagrees. It also asserts that the README names no harness absent from `INTEGRATION_TYPES`, which is what makes the twelve-harness claim checkable by the engineer most likely to check it. (#731)
+
+- Stop the harness copy folding the Python SDK into the twelve. The twelve are TWO classes — ten coding CLIs and two chat/assistant gateways — and an SDK-instrumented agent is a separate door, not a third class of the twelve. Worse, the copy said "same events, same policies" of all three: the SDK reports events, it does not sit in the tool-call path, so it delivers observability, sessions and audits **without** enforcement, exactly as `reference/python-sdk` has always said. Blocking an unsafe action on that path needs a hook at the runtime's tool boundary. The drift guard now asserts both the two-class split and that no copy describes the SDK as carrying policies. (#731)
+
+- Point the fourteen translation links at the tree the translation pipeline actually maintains. They resolved to `docs-old/i18n/`, frozen when the Mintlify site replaced it, while `.github/workflows/translate-docs.yml` writes `docs/i18n/` — so every non-English reader got a README that would never be updated again, and the two sets had already diverged. The drift guard asserts the link target. (#731)
 
 ### Fixes
 
+- Stop the local audit report claiming it would have blocked things it only detected. `totalHits` folds in `detectorRows` (source `audit-detector`) — replay-only findings with no builtin behind them, whose own fix column renders `audit-only` and which the slipping-through section already labels "caught by audit, not blocked" — yet the TL;DR line users paste publicly said the agent did N things "`failproofai` would have stopped". "Caught" is true of both halves; "would have stopped" was true of one. The same footer also named six integrations when `runAuditInner` defaults to all twelve. (#731)
+
 - Stop the localized navigation referencing pages that were never translated, which is what still discarded a partial run. `--allow-partial` published what succeeded — and then `--update-nav` regenerated the nav from the ENGLISH tree, emitting an entry for the failed page in the language that failed it, so `mintlify validate` rejected the missing file and the job died before its push anyway. The 784 pages that HAD translated went with it, which is precisely the loss `--allow-partial` exists to prevent. Nav generation now omits any localized page whose file is not on disk, prunes a group left with no pages and a tab left with no groups, and keeps an `openapi` group that never had pages to begin with. The check is injected rather than hardcoded, so the pure transform stays testable and the two paths that actually write `docs.json` get the real one. This also closes the same hazard from every other direction it can arrive from — a pruned page, or a translation that only exists on an unmerged branch — because the nav is now derived from what is present rather than from what English says should be. (#725)
 
 - Stop the localized nav crashing on a group that has no pages, and stop it dropping the properties it does have. `buildLanguageNav` rebuilt every group as `{group, pages}` and called `group.pages.map(...)` unconditionally. The docs rebuild added `{group, expanded, openapi}` — a group whose content is an OpenAPI spec and has no pages at all — so `--update-nav` died with `TypeError: undefined is not an object`, **after 784 pages had already been translated**, taking the whole nightly run with it for the second night running. Groups are now rebuilt by spreading the English group, so `expanded`, `icon` and `openapi` survive instead of being silently discarded from every non-English nav; a group with no pages is carried through untouched (the spec is not translated, and dropping it would remove the API reference from thirteen languages); and `pages` entries that are themselves nested groups recurse rather than being prefixed as if they were paths. `pages` is optional on the type now, which is what it always was in the data. (#725)
diff --git a/README.md b/README.md
index 948371c47..ededeb50c 100644
--- a/README.md
+++ b/README.md
@@ -11,11 +11,13 @@
 [![Docs](https://img.shields.io/badge/docs-befailproof.ai-002CA7?style=flat-square)](https://docs.befailproof.ai/)
 [![License](https://img.shields.io/badge/license-MIT%20%2B%20Commons%20Clause-blue?style=flat-square)](./LICENSE)
 
-**Translations:** [简体中文](./docs-old/i18n/README.zh.md) · [日本語](./docs-old/i18n/README.ja.md) · [한국어](./docs-old/i18n/README.ko.md) · [Español](./docs-old/i18n/README.es.md) · [Português](./docs-old/i18n/README.pt-br.md) · [Deutsch](./docs-old/i18n/README.de.md) · [Français](./docs-old/i18n/README.fr.md) · [Руссий](./docs-old/i18n/README.ru.md) · [हिन्दी](./docs-old/i18n/README.hi.md) · [Türkçe](./docs-old/i18n/README.tr.md) · [Tiếng Việt](./docs-old/i18n/README.vi.md) · [Italiano](./docs-old/i18n/README.it.md) · [العربية](./docs-old/i18n/README.ar.md) · [עברית](./docs-old/i18n/README.he.md)
+**Translations:** [简体中文](./docs/i18n/README.zh.md) · [日本語](./docs/i18n/README.ja.md) · [한국어](./docs/i18n/README.ko.md) · [Español](./docs/i18n/README.es.md) · [Português](./docs/i18n/README.pt-br.md) · [Deutsch](./docs/i18n/README.de.md) · [Français](./docs/i18n/README.fr.md) · [Русский](./docs/i18n/README.ru.md) · [हिन्दी](./docs/i18n/README.hi.md) · [Türkçe](./docs/i18n/README.tr.md) · [Tiếng Việt](./docs/i18n/README.vi.md) · [Italiano](./docs/i18n/README.it.md) · [العربية](./docs/i18n/README.ar.md) · [עברית](./docs/i18n/README.he.md)
 
-**Runtime failure resolution for coding agents.**
-Hooks into Claude Code and Codex. Catches loops, dangerous actions, and secret leaks
-before they become incidents. Zero latency. Runs locally.
+**Observability and enforcement for every harness your agents run in.**
+Wherever your agents run, we see it — and we can say no. Failproof hooks 12 agent
+harnesses — coding CLIs like Claude Code and Codex, chat gateways like Hermes,
+self-hosted assistants like OpenClaw — capturing every run and blocking dangerous
+tool calls before they execute. 40 built-in policies. Zero latency. Runs locally.
 
 </div>
 
@@ -25,7 +27,15 @@ before they become incidents. Zero latency. Runs locally.
 
 ---
 
-## Supported agent CLIs
+## Supported harnesses
+
+Twelve harnesses in two classes — ten coding CLIs, and two chat and assistant
+gateways (Hermes, OpenClaw). Same events, same policies, same session history,
+whichever one your agent runs in.
+
+Agents that run in none of them report through the [Python SDK](https://docs.befailproof.ai/reference/python-sdk),
+which gives you tracing, sessions and audits. Enforcement there needs a hook in
+your own runtime — [talk to us](mailto:support@befailproof.ai) and we'll map it.
 
 <!-- A 6-column table instead of inline <img> runs: table columns never re-wrap,
      so the grid stays 2×6 at any window width (scrolling on very narrow screens
@@ -129,7 +139,7 @@ failproofai policies --install   # or just run `failproofai` and accept the firs
 failproofai
 ```
 
-30 built-in policies activate immediately. Dashboard at `localhost:8020`. Disable the first-run prompt with `FAILPROOFAI_NO_FIRST_RUN=1`.
+40 built-in policies activate immediately. Dashboard at `localhost:8020`. Disable the first-run prompt with `FAILPROOFAI_NO_FIRST_RUN=1`.
 
 ---
 
@@ -137,13 +147,19 @@ failproofai
 
 | Policy | What it blocks |
 |---|---|
-| `block-push-master` | Direct pushes to `main` / `master` |
-| `block-force-push` | `git push --force` |
-| `block-work-on-main` | Commits, merges, rebases on `main` / `master` |
+| `sanitize-api-keys` | API keys leaking into the agent's context |
+| `block-env-files` | Reads of `.env` and other secret files |
+| `warn-repeated-tool-calls` | The agent looping on the same call |
+| `block-sudo` | Privilege escalation |
+| `warn-destructive-sql` | `DROP`, `TRUNCATE`, unbounded `DELETE` |
+| `block-terraform` / `block-kubectl` | Unreviewed changes to live infrastructure |
 | `block-rm-rf` | Recursive file deletion |
-| `sanitize-api-keys` | API keys leaking into agent context |
+| `block-force-push` / `block-push-master` | `git push --force`, direct pushes to `main` |
 
-→ [All 30 built-in policies](https://docs.befailproof.ai/policies/builtin)
+The first five apply to any agent that can call a tool. The last three are the
+developer favourites — coding CLIs are the harness class we cover deepest.
+
+→ [All 40 built-in policies](https://docs.befailproof.ai/policies/builtin)
 
 ---
 
@@ -178,24 +194,61 @@ Three decisions available to every policy:
 
 ---
 
-## Session visibility
+## Observability
+
+Enforcement is one half. The other half is seeing what the agent actually did.
+
+Run `failproofai` with no arguments and it serves a dashboard on `localhost:8020`
+reading the run history already on your machine — no account, no signup, nothing
+leaving the box. You get the session list, the sequence of model calls, tool calls
+and hook decisions inside each run, what was blocked and what the policy told the
+agent, and an offline audit (`failproofai audit`) that scans your history for risky
+patterns and suggests policies to stop them.
+
+→ [Local dashboard](https://docs.befailproof.ai/reference/local-dashboard) ·
+[Read a trace](https://docs.befailproof.ai/sessions/read-a-trace) ·
+[Local audit](https://docs.befailproof.ai/audits/local-audit)
+
+**Failproof AI Observability** is the hosted side of the same data model, for teams
+running agents across a fleet: every run from every harness in one place, an
+execution graph with parallel sub-agents on their own lanes, p50/p95/p99 latency
+for models, tools and hooks, per-model cost and context-window tracking, error
+tracking, SQL over your own traces with shareable dashboards, evaluations scored by
+your own service, scheduled audits that turn recurring failures into evidence-backed
+findings, and alerts routed to Slack, email or a signed webhook. Self-hosting in your
+own cluster is available on the Enterprise plan.
 
-Every tool call your agent makes is logged locally. The dashboard shows what ran,
-what was blocked, and what the policy told the agent — so you're not guessing
-when something goes wrong. → [Dashboard guide](https://docs.befailproof.ai/sessions/overview)
+→ [Sessions](https://docs.befailproof.ai/sessions/overview) ·
+[Audits](https://docs.befailproof.ai/audits/overview) ·
+[Book a demo](https://befailproof.ai/get-a-demo)
 
 ---
 
 ## Documentation
 
-| | |
+| Start | |
 |---|---|
-| [Getting Started](https://docs.befailproof.ai/start/quickstart) | Installation and first steps |
-| [Built-in Policies](https://docs.befailproof.ai/policies/builtin) | All 30 policies with parameters |
-| [Custom Policies](https://docs.befailproof.ai/policies/custom) | Write your own |
+| [Quickstart](https://docs.befailproof.ai/start/quickstart) | Install, connect a harness, see the first run |
+| [Concepts](https://docs.befailproof.ai/start/concepts) | How the hook system works |
+| [Supported harnesses](https://docs.befailproof.ai/reference/harnesses) | All 12, and what each one can enforce |
+
+| Observe | |
+|---|---|
+| [Sessions](https://docs.befailproof.ai/sessions/overview) | Follow a run: models, tools, errors, latency |
+| [Read a trace](https://docs.befailproof.ai/sessions/read-a-trace) | What the execution graph is telling you |
+| [Audits](https://docs.befailproof.ai/audits/overview) | Find failure patterns across many sessions |
+| [Local dashboard](https://docs.befailproof.ai/reference/local-dashboard) | `localhost:8020`, no account needed |
+
+| Enforce | |
+|---|---|
+| [Built-in policies](https://docs.befailproof.ai/policies/builtin) | All 40 policies with parameters |
+| [Custom policies](https://docs.befailproof.ai/policies/custom) | Write your own |
 | [Configuration](https://docs.befailproof.ai/policies/local-configuration) | Config scopes and merge rules |
-| [Dashboard](https://docs.befailproof.ai/sessions/overview) | Session monitor and policy activity |
-| [Architecture](https://docs.befailproof.ai/start/concepts) | How the hook system works |
+
+| Instrument your own agent | |
+|---|---|
+| [Python SDK](https://docs.befailproof.ai/reference/python-sdk) | Report runs from an agent with no harness |
+| [Policy SDK](https://docs.befailproof.ai/reference/policy-sdk) | `allow` / `deny` / `instruct` reference |
 
 ---
 
diff --git a/SECURITY.md b/SECURITY.md
index 24ba3dc7b..ecf874f53 100644
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -1,9 +1,10 @@
 # Security Policy
 
-failproofai is an npm package that installs hooks into AI coding agents (Claude
-Code, Codex, and others) and runs locally in developers' environments, observing
-and gating agent actions. A compromised dependency would therefore run on our
-users' machines, inside their agent sessions — so we treat our supply chain as
+failproofai is an npm package that installs hooks into AI agent harnesses
+(Claude Code, Codex, Cursor, the Hermes and OpenClaw gateways, and others) and
+runs locally in the environment the agent runs in, observing and gating agent
+actions. A compromised dependency would therefore run on our
+users' machines and servers, inside their agent sessions — so we treat our supply chain as
 part of our users' trust boundary. This document covers how to report a
 vulnerability and how our dependency supply-chain scanning works.
 
diff --git a/__tests__/scripts/copy-counts.test.ts b/__tests__/scripts/copy-counts.test.ts
new file mode 100644
index 000000000..0a035ce22
--- /dev/null
+++ b/__tests__/scripts/copy-counts.test.ts
@@ -0,0 +1,168 @@
+// @vitest-environment node
+/**
+ * Drift guard for the NUMBERS AND HARNESS NAMES in user-facing prose.
+ *
+ * The positioning rewrite put four counts into copy read by nearly everyone who
+ * meets this project — 12 harnesses, 40 built-in policies, 29 canonical events,
+ * and blocking verified on 12/12 PreToolUse and 8/12 Stop. Those counts are only
+ * worth printing flat, without hedging, because `enforcement-capability.ts` is a
+ * machine-readable matrix with the probed version on every row. A stale count
+ * turns that asset into a liability: an engineer who checks one number and finds
+ * it wrong has a reason to discount the whole claim.
+ *
+ * This is not hypothetical. Before this file existed, README.md advertised "30
+ * built-in policies" in three places and the landing site's schema.org block said
+ * 39 — while the real count was 40, and had been for some time. Both were
+ * hand-maintained, nothing read them, so both rotted silently. Same failure mode
+ * __tests__/hooks/dogfood-configs.test.ts was written to catch for the dogfood
+ * configs.
+ *
+ * It also guards the two claims most likely to be caught out by a reader: that
+ * the twelve are TWO classes (ten coding CLIs, two gateways) with the Python SDK
+ * as a separate door rather than a third class, and that the SDK is never
+ * described as carrying policies — it reports events, it does not sit in the
+ * tool-call path, so it observes without enforcing.
+ *
+ * SCOPE: this repo only. The landing site (befailproof.ai) lives in a different
+ * repo and carries the same counts in app/layout.js and app/llms.txt/route.js;
+ * it needs its own check. See FailproofAI/agenteye#629.
+ */
+import { describe, it, expect } from "vitest";
+import { existsSync, readFileSync } from "node:fs";
+import { resolve } from "node:path";
+import { INTEGRATION_TYPES, HOOK_EVENT_TYPES } from "@/src/hooks/types";
+import { ENFORCEMENT_CAPABILITY } from "@/src/hooks/enforcement-capability";
+import { BUILTIN_POLICIES } from "@/src/hooks/builtin-policies";
+
+const ROOT = process.cwd();
+const read = (p: string) => readFileSync(resolve(ROOT, p), "utf8");
+
+/** Every count asserted below, derived from source rather than restated. */
+const TRUTH = {
+  harnesses: INTEGRATION_TYPES.length,
+  policies: BUILTIN_POLICIES.length,
+  events: HOOK_EVENT_TYPES.length,
+  preToolUseBlocks: INTEGRATION_TYPES.filter(
+    (c) => ENFORCEMENT_CAPABILITY[c]?.PreToolUse === "block",
+  ).length,
+  stopBlocks: INTEGRATION_TYPES.filter((c) => ENFORCEMENT_CAPABILITY[c]?.Stop === "block").length,
+};
+
+/**
+ * Files that state a count in prose a reader acts on. Deliberately short: this
+ * guards the strings with the most readers, not every mention in the tree.
+ */
+const COPY_FILES = ["README.md", "package.json", "docs/index.mdx"];
+
+describe("copy counts match source", () => {
+  it("the derived counts are what the copy claims", () => {
+    // Fails loudly if someone adds a harness or a policy without re-reading the
+    // copy. Update this block ONLY together with every string it guards.
+    expect(TRUTH).toEqual({
+      harnesses: 12,
+      policies: 40,
+      events: 29,
+      preToolUseBlocks: 12,
+      stopBlocks: 8,
+    });
+  });
+
+  it.each(COPY_FILES)("%s states no stale policy count", (file) => {
+    const text = read(file);
+    const claimed = [...text.matchAll(/(\d+)\s+built-in polic/gi)].map((m) => Number(m[1]));
+    for (const n of claimed) expect(n).toBe(TRUTH.policies);
+  });
+
+  it.each(COPY_FILES)("%s states no stale harness count", (file) => {
+    const text = read(file);
+    const claimed = [...text.matchAll(/(\d+)\s+(?:agent\s+)?harness(?:es)?\b/gi)].map((m) =>
+      Number(m[1]),
+    );
+    for (const n of claimed) expect(n).toBe(TRUTH.harnesses);
+  });
+
+  it("README names only harnesses that ship", () => {
+    const text = read("README.md").toLowerCase();
+    // Named on marketing surfaces but absent from INTEGRATION_TYPES. Keeping
+    // them out of the README is what makes the harness list checkable.
+    for (const absent of ["gemini cli", "langgraph", "deep agents"]) {
+      expect(text).not.toContain(absent);
+    }
+  });
+
+  it("README names the two harnesses that prove the product is not coding-only", () => {
+    // The whole positioning rests on the harness list spanning classes. Hermes
+    // (Slack/Telegram gateway) and OpenClaw (self-hosted assistant) are the two
+    // shipped integrations that were never coding tools; if a future edit drops
+    // them from the hero, the breadth claim silently loses its evidence.
+    const text = read("README.md").toLowerCase();
+    expect(text).toContain("hermes");
+    expect(text).toContain("openclaw");
+  });
+
+  it("the translated READMEs are linked from the tree the pipeline maintains", () => {
+    // docs-old/i18n/ was frozen when the Mintlify site replaced it; the
+    // translation workflow writes docs/i18n/. Linking the frozen copies means
+    // non-English readers get a README that is never updated again.
+    expect(read("README.md")).not.toContain("docs-old/i18n/");
+  });
+
+  it("every translation link resolves to a file on disk", () => {
+    // The bug this guards was a link pointing at a real file in the WRONG tree,
+    // so "not docs-old" alone would not have caught a typo'd locale either.
+    const links = [...read("README.md").matchAll(/\]\((\.\/docs\/i18n\/[^)]+)\)/g)].map(
+      (m) => m[1],
+    );
+    expect(links.length).toBe(14);
+    for (const href of links) {
+      expect({ href, exists: existsSync(resolve(ROOT, href)) }).toEqual({ href, exists: true });
+    }
+  });
+
+  it("README names every shipped harness, in the class it belongs to", () => {
+    // The twelve are TWO classes: ten coding CLIs and two gateways. The Python
+    // SDK is a separate door, not a third class of the twelve — and it reports
+    // events rather than sitting in the tool-call path, so it observes without
+    // enforcing. Copy that folds it in overstates what it does, which is the
+    // single easiest way for a reader to catch us out.
+    const text = read("README.md");
+    expect(text).toMatch(/Twelve harnesses in two classes/);
+    expect(text).toMatch(/ten coding CLIs/);
+    expect(text).toMatch(/Hermes, OpenClaw/);
+    // The SDK must never be described as carrying policies.
+    const sdkClaim = /Python SDK[^.]*same polic/i;
+    expect(sdkClaim.test(text)).toBe(false);
+  });
+
+  it("the harnesses page agrees with the source list, harness for harness", () => {
+    const page = read("docs/reference/harnesses.mdx");
+    const NAMES: Record<(typeof INTEGRATION_TYPES)[number], string> = {
+      claude: "Claude Code",
+      codex: "Codex",
+      copilot: "GitHub Copilot CLI",
+      cursor: "Cursor",
+      opencode: "OpenCode",
+      pi: "Pi",
+      hermes: "Hermes",
+      openclaw: "OpenClaw",
+      factory: "Factory Droid",
+      devin: "Devin CLI",
+      antigravity: "Antigravity CLI",
+      goose: "Goose",
+    };
+    // Adding an integration without documenting it is the drift this catches.
+    expect(Object.keys(NAMES).sort()).toEqual([...INTEGRATION_TYPES].sort());
+    for (const display of Object.values(NAMES)) expect(page).toContain(display);
+  });
+
+  it("the enforcement split stated in the docs matches the capability matrix", () => {
+    // README and quickstart print these flat, without hedging, which is only
+    // defensible because enforcement-capability.ts carries the probed version on
+    // every row. If a re-probe moves a row, the prose has to move with it.
+    const quickstart = read("docs/start/quickstart.mdx");
+    expect(quickstart).toContain("verified on all 12");
+    expect(TRUTH.preToolUseBlocks).toBe(12);
+    expect(quickstart).toMatch(/verified on 8\b/);
+    expect(TRUTH.stopBlocks).toBe(8);
+  });
+});
diff --git a/app/audit/_components/share-templates.ts b/app/audit/_components/share-templates.ts
index aa5b18448..190762e6f 100644
--- a/app/audit/_components/share-templates.ts
+++ b/app/audit/_components/share-templates.ts
@@ -34,49 +34,49 @@ const PASTE_LINE = "[ audit card copied to your clipboard — paste it into the
 /** Short, curiosity-forward — for X / Twitter. Ends on the npx CTA + @failproofai. */
 export const X_TEMPLATES: ((c: ShareCtx) => string)[] = [
   ({ score, arch }) =>
-    `my coding agent has a personality and it's ${arch}. did not see that coming.\n\nran a failproof audit, scored ${score}/100, all local. find out what yours is → ${X_CTA}`,
+    `my agent has a personality and it's ${arch}. did not see that coming.\n\nran a failproof audit, scored ${score}/100, all local. find out what yours is → ${X_CTA}`,
   ({ score, arch }) =>
-    `spent 30 seconds auditing my coding agent. learned more about it than i did all month.\n\nit's ${arch}. scored ${score}/100. → ${X_CTA}`,
+    `spent 30 seconds auditing my agent. learned more about it than i did all month.\n\nit's ${arch}. scored ${score}/100. → ${X_CTA}`,
   ({ score }) =>
     `my agent scored ${score}/100. think yours can beat it?\n\n30 sec audit, runs locally, gives you a personality too → ${X_CTA}`,
   ({ arch }) =>
-    `turns out my coding agent is ${arch} and honestly it explains everything.\n\n→ ${X_CTA}`,
+    `turns out my agent is ${arch} and honestly it explains everything.\n\n→ ${X_CTA}`,
   ({ score, arch }) =>
     `ran a failproof audit on my agent. it reads how it actually behaves when things break, not just the happy path.\n\npersonality: ${arch}. score: ${score}/100. all local → ${X_CTA}`,
   ({ score, arch }) =>
-    `got to know my coding agent today. it's ${arch} and scored ${score}/100. wow.\n\naudit yours in 30s, nothing leaves your machine → ${X_CTA}`,
+    `got to know my agent today. it's ${arch} and scored ${score}/100. wow.\n\naudit yours in 30s, nothing leaves your machine → ${X_CTA}`,
   ({ score, arch }) =>
-    `your coding agent has a personality. you just haven't met it.\n\ni met mine today: ${arch}, ${score}/100. → ${X_CTA}`,
+    `your agent has a personality. you just haven't met it.\n\ni met mine today: ${arch}, ${score}/100. → ${X_CTA}`,
   ({ score, arch }) =>
     `everyone talks about what their agent can build. nobody talks about how it acts when it breaks.\n\nmine's ${arch}, scored ${score}/100, all local → ${X_CTA}`,
   ({ score, arch }) =>
     `every agent builder should run this once.\n\n30 sec, local, gives your agent a personality and a quality score. mine: ${arch}, ${score}/100.\n\n${X_CTA}`,
   ({ score, arch }) =>
-    `i'll go first: ${arch}, ${score}/100.\n\nwhat's your coding agent? 30 sec audit, no signup → ${X_CTA}`,
+    `i'll go first: ${arch}, ${score}/100.\n\nwhat's your agent? 30 sec audit, no signup → ${X_CTA}`,
 ];
 
 /** Longer, reflective — for LinkedIn. Ends on the npx CTA + @Failproof AI. */
 export const LI_TEMPLATES: ((c: ShareCtx) => string)[] = [
   ({ score, arch }) =>
-    `I've spent more hours with my coding agent this month than with most people I know. Today I realized I couldn't tell you a single thing about how it actually behaves.\n\nSo I audited it. Turns out it's ${arch}, and it scored ${score}/100.\n\nThe audit reads its real history, including how it handles failures, not just the clean runs. 30 seconds, runs entirely locally.\n\nMeet yours: ${LI_CTA}`,
+    `I've spent more hours with my agent this month than with most people I know. Today I realized I couldn't tell you a single thing about how it actually behaves.\n\nSo I audited it. Turns out it's ${arch}, and it scored ${score}/100.\n\nThe audit reads its real history, including how it handles failures, not just the clean runs. 30 seconds, runs entirely locally.\n\nMeet yours: ${LI_CTA}`,
   ({ score, arch }) =>
-    `Took a personality test today. It wasn't for me — it was for my AI coding agent.\n\nThe result: ${arch}, with a quality score of ${score}/100.\n\nIt works by reading the agent's actual run history, so the personality comes from how it really behaves, not a vibe. Took 30 seconds and stayed local.\n\nTry it on yours: ${LI_CTA}`,
+    `Took a personality test today. It wasn't for me — it was for my AI agent.\n\nThe result: ${arch}, with a quality score of ${score}/100.\n\nIt works by reading the agent's actual run history, so the personality comes from how it really behaves, not a vibe. Took 30 seconds and stayed local.\n\nTry it on yours: ${LI_CTA}`,
   ({ score, arch }) =>
     `Everyone asks what their AI agent can build. I just found out mine has a temper.\n\nAfter watching how it reacts to a failed command, the audit called it ${arch}. It also scored its quality at ${score}/100.\n\nOddly useful to see it written down. 30 seconds, all local: ${LI_CTA}`,
   ({ score, arch }) =>
     `My agent does this very specific thing every time a command fails. Today I learned there's basically a name for it.\n\nRan a quick audit and it came back ${arch}, scored ${score}/100. The whole read is based on how the agent handles things going wrong, which is where its real character shows.\n\n30 seconds, nothing leaves your machine: ${LI_CTA}`,
   ({ score, arch }) =>
-    `My AI coding agent scored ${score}/100 today. I didn't know you could even measure that.\n\nThe same audit also handed it a personality: ${arch}.\n\nIt reads the agent's real history and tells you where it's solid and where it slips. Took 30 seconds and ran fully locally: ${LI_CTA}`,
+    `My AI agent scored ${score}/100 today. I didn't know you could even measure that.\n\nThe same audit also handed it a personality: ${arch}.\n\nIt reads the agent's real history and tells you where it's solid and where it slips. Took 30 seconds and ran fully locally: ${LI_CTA}`,
   ({ score, arch }) =>
-    `I'm usually the first to scroll past "audit your X" tools. This one took 30 seconds and actually told me something.\n\nMy coding agent is ${arch}, and it scored ${score}/100.\n\nIt reads the agent's real run history rather than asking me anything, and none of it leaves the machine: ${LI_CTA}`,
+    `I'm usually the first to scroll past "audit your X" tools. This one took 30 seconds and actually told me something.\n\nMy agent is ${arch}, and it scored ${score}/100.\n\nIt reads the agent's real run history rather than asking me anything, and none of it leaves the machine: ${LI_CTA}`,
   ({ score, arch }) =>
-    `What is your coding agent's personality? I realized today that I had no answer.\n\nSo I checked. Mine is ${arch}, scored ${score}/100, with a clear read on where it shines and where it slips.\n\nIf you work with agents, run it and drop yours in the comments. 30 seconds, fully local: ${LI_CTA}`,
+    `What is your agent's personality? I realized today that I had no answer.\n\nSo I checked. Mine is ${arch}, scored ${score}/100, with a clear read on where it shines and where it slips.\n\nIf you work with agents, run it and drop yours in the comments. 30 seconds, fully local: ${LI_CTA}`,
   ({ score, arch }) =>
-    `I trusted my coding agent a little less after this morning, and quite a bit more by the afternoon.\n\nI ran an audit on it. It showed me exactly where it's reliable and where it isn't, gave it a personality (${arch}), and scored it ${score}/100.\n\nKnowing the gaps is what made me trust it more. 30 seconds, all local: ${LI_CTA}`,
+    `I trusted my agent a little less after this morning, and quite a bit more by the afternoon.\n\nI ran an audit on it. It showed me exactly where it's reliable and where it isn't, gave it a personality (${arch}), and scored it ${score}/100.\n\nKnowing the gaps is what made me trust it more. 30 seconds, all local: ${LI_CTA}`,
   ({ score, arch }) =>
     `I almost skipped this because I assumed it would quietly ship my code off somewhere. It doesn't. Everything runs locally.\n\n30 seconds later I had my agent's personality (${arch}) and a quality score (${score}/100).\n\nIf privacy is usually what stops you from trying these, this one is different: ${LI_CTA}`,
   ({ score, arch }) =>
-    `Found out my AI coding agent has a personality type. Found out it's ${arch}. Found out that explains a lot.\n\nIt scored ${score}/100 too. The audit reads how the agent handles failures, not just the runs where everything goes fine.\n\n30 seconds, runs locally: ${LI_CTA}`,
+    `Found out my AI agent has a personality type. Found out it's ${arch}. Found out that explains a lot.\n\nIt scored ${score}/100 too. The audit reads how the agent handles failures, not just the runs where everything goes fine.\n\n30 seconds, runs locally: ${LI_CTA}`,
 ];
 
 /** djb2 hash — stable per seed so the template choice is deterministic. */
diff --git a/app/policies/hooks-client.tsx b/app/policies/hooks-client.tsx
index 3b0422be5..4761ed520 100644
--- a/app/policies/hooks-client.tsx
+++ b/app/policies/hooks-client.tsx
@@ -1951,7 +1951,7 @@ export default function HooksClient({ initialTab = "activity" }: { initialTab?:
               margin: "0 0 24px",
             }}
           >
-            switch policies on or off across your installed agent CLIs.
+            switch policies on or off across your installed harnesses.
           </p>
         )}
 
diff --git a/docs/docs.json b/docs/docs.json
index 674dc79a2..87bfb98e9 100644
--- a/docs/docs.json
+++ b/docs/docs.json
@@ -71,6 +71,7 @@
                 "pages": [
                   "index",
                   "start/quickstart",
+                  "reference/harnesses",
                   "sessions/assistant",
                   "start/first-audit",
                   "start/first-policy",
@@ -208,6 +209,7 @@
                 "pages": [
                   "zh/index",
                   "zh/start/quickstart",
+                  "zh/reference/harnesses",
                   "zh/sessions/assistant",
                   "zh/start/first-audit",
                   "zh/start/first-policy",
@@ -345,6 +347,7 @@
                 "pages": [
                   "ja/index",
                   "ja/start/quickstart",
+                  "ja/reference/harnesses",
                   "ja/sessions/assistant",
                   "ja/start/first-audit",
                   "ja/start/first-policy",
@@ -482,6 +485,7 @@
                 "pages": [
                   "ko/index",
                   "ko/start/quickstart",
+                  "ko/reference/harnesses",
                   "ko/sessions/assistant",
                   "ko/start/first-audit",
                   "ko/start/first-policy",
@@ -619,6 +623,7 @@
                 "pages": [
                   "es/index",
                   "es/start/quickstart",
+                  "es/reference/harnesses",
                   "es/sessions/assistant",
                   "es/start/first-audit",
                   "es/start/first-policy",
@@ -756,6 +761,7 @@
                 "pages": [
                   "pt-br/index",
                   "pt-br/start/quickstart",
+                  "pt-br/reference/harnesses",
                   "pt-br/sessions/assistant",
                   "pt-br/start/first-audit",
                   "pt-br/start/first-policy",
@@ -893,6 +899,7 @@
                 "pages": [
                   "de/index",
                   "de/start/quickstart",
+                  "de/reference/harnesses",
                   "de/sessions/assistant",
                   "de/start/first-audit",
                   "de/start/first-policy",
@@ -1030,6 +1037,7 @@
                 "pages": [
                   "fr/index",
                   "fr/start/quickstart",
+                  "fr/reference/harnesses",
                   "fr/sessions/assistant",
                   "fr/start/first-audit",
                   "fr/start/first-policy",
@@ -1167,6 +1175,7 @@
                 "pages": [
                   "ru/index",
                   "ru/start/quickstart",
+                  "ru/reference/harnesses",
                   "ru/sessions/assistant",
                   "ru/start/first-audit",
                   "ru/start/first-policy",
@@ -1304,6 +1313,7 @@
                 "pages": [
                   "hi/index",
                   "hi/start/quickstart",
+                  "hi/reference/harnesses",
                   "hi/sessions/assistant",
                   "hi/start/first-audit",
                   "hi/start/first-policy",
@@ -1441,6 +1451,7 @@
                 "pages": [
                   "tr/index",
                   "tr/start/quickstart",
+                  "tr/reference/harnesses",
                   "tr/sessions/assistant",
                   "tr/start/first-audit",
                   "tr/start/first-policy",
@@ -1578,6 +1589,7 @@
                 "pages": [
                   "vi/index",
                   "vi/start/quickstart",
+                  "vi/reference/harnesses",
                   "vi/sessions/assistant",
                   "vi/start/first-audit",
                   "vi/start/first-policy",
@@ -1715,6 +1727,7 @@
                 "pages": [
                   "it/index",
                   "it/start/quickstart",
+                  "it/reference/harnesses",
                   "it/sessions/assistant",
                   "it/start/first-audit",
                   "it/start/first-policy",
@@ -1852,6 +1865,7 @@
                 "pages": [
                   "ar/index",
                   "ar/start/quickstart",
+                  "ar/reference/harnesses",
                   "ar/sessions/assistant",
                   "ar/start/first-audit",
                   "ar/start/first-policy",
@@ -1989,6 +2003,7 @@
                 "pages": [
                   "he/index",
                   "he/start/quickstart",
+                  "he/reference/harnesses",
                   "he/sessions/assistant",
                   "he/start/first-audit",
                   "he/start/first-policy",
diff --git a/docs/index.mdx b/docs/index.mdx
index 9741f4b89..45985b74b 100644
--- a/docs/index.mdx
+++ b/docs/index.mdx
@@ -1,11 +1,13 @@
 ---
 title: "Make your agent failproof"
-description: "Failproof AI traces agent behavior, audits sessions for failures, and deploys policies that stop repeat incidents."
+description: "Observability and enforcement for every harness your agents run in — coding CLIs, chat gateways, self-hosted assistants, and your own instrumented agents."
 icon: "shield-check"
 ---
 
 Failproof AI helps teams understand what agents did, find where they failed, and deploy safeguards before the same behavior happens again.
 
+A **harness** is whatever your agent actually runs inside. Failproof AI hooks 12 of them — coding CLIs like Claude Code and Codex, chat gateways like Hermes, self-hosted assistants like OpenClaw — and the same events, the same policies, and the same session history apply to every one. Agents with no harness report in through the [Python SDK](/reference/python-sdk), which traces and audits them; enforcing a policy there needs a hook in your own runtime.
+
 <Columns cols={2}>
   <Card title="Set Failproof up" icon="wand-sparkles" href="/start/quickstart">
     Use the skill to instrument your project, connect it, and verify that agent logs arrive.
diff --git a/docs/reference/harnesses.mdx b/docs/reference/harnesses.mdx
index e38287e1b..f8b397c50 100644
--- a/docs/reference/harnesses.mdx
+++ b/docs/reference/harnesses.mdx
@@ -1,10 +1,17 @@
 ---
 title: "Agent harnesses"
-description: "Capture sessions and enforce policies in supported agent CLIs."
+description: "Capture sessions and enforce policies across all 12 supported agent harnesses."
 icon: "plug-zap"
 ---
 
-Failproof AI supports Claude Code, Codex, GitHub Copilot CLI, Cursor, OpenCode, Pi, Hermes, OpenClaw, Factory Droid, Devin CLI, Antigravity CLI, and Goose.
+A harness is whatever your agent actually runs inside. Failproof AI supports twelve of them, in two classes:
+
+- **Coding CLIs** (10) — Claude Code, Codex, GitHub Copilot CLI, Cursor, OpenCode, Pi, Factory Droid, Devin CLI, Antigravity CLI, Goose
+- **Chat and assistant gateways** (2) — Hermes (Slack, Telegram, cron), OpenClaw (self-hosted assistant)
+
+The same policies and the same session history apply whichever one an agent runs in. One adapter layer maps each harness's native event names, tool names, and tool-input fields onto 29 canonical events before any policy runs.
+
+An agent that runs in **none** of the twelve is instrumented directly with the [Python SDK](/reference/python-sdk). That is a different contract, and worth stating plainly: the SDK delivers tracing, sessions, evaluations and audits — **it does not enforce policies on its own.** Blocking an unsafe action before it executes needs an enforcement hook at your runtime's tool boundary; [contact us](mailto:support@befailproof.ai) and we will map it.
 
 | Harness | Supported hook scopes |
 | --- | --- |
diff --git a/docs/sessions/live-events.mdx b/docs/sessions/live-events.mdx
index 77e1e95cd..65d93a686 100644
--- a/docs/sessions/live-events.mdx
+++ b/docs/sessions/live-events.mdx
@@ -15,7 +15,7 @@ Live events help you confirm instrumentation and watch a risky run without waiti
     3. Filter by environment, event type, agent, or session. Use search for payload text.
     4. Select an event to inspect its summary and details. Follow its session link for the full trace.
 
-    ![The live Events stream with time, environment, event type, agent, model, context, session, and text filters.](/images/dashboard/events-stream-current.png)
+    ![The live Events stream showing traffic from a Hermes chat gateway alongside coding-CLI runs, filtered by time, environment, event type, agent, model, context, and session.](/images/dashboard/events-stream-current.png)
   </Tab>
   <Tab title="CLI">
     ```bash
@@ -41,3 +41,5 @@ Event types include agent lifecycle events, model requests and responses, tool u
 </Tip>
 
 If no events appear, run `failproofai config --status`, then [troubleshoot ingestion](/reference/troubleshooting).
+
+Agents that do not run in one of the 12 supported [harnesses](/reference/harnesses) report the same event types through the [Python SDK](/reference/python-sdk), including the human-in-the-loop events (`human_wait`, `human_input`, `human_interrupt`) that gateway and production agents depend on.
diff --git a/docs/sessions/overview.mdx b/docs/sessions/overview.mdx
index 8cbbd76bf..c03551eb1 100644
--- a/docs/sessions/overview.mdx
+++ b/docs/sessions/overview.mdx
@@ -6,6 +6,8 @@ icon: "workflow"
 
 A session is the best starting point when an agent behaves unexpectedly. It joins the model requests, responses, tool calls, human interactions, errors, evaluations, and policy decisions that belong to one run.
 
+Sessions look the same whichever harness produced them. A Claude Code run rewriting a repository, a Hermes agent answering a customer in Slack, and a Python service instrumented with the SDK all land in the same trace format, so one view covers the fleet.
+
 <div style={{ position: "relative", width: "100%", paddingBottom: "56.25%", height: 0, overflow: "hidden", borderRadius: "12px", margin: "1.5rem 0" }}>
   <iframe src="https://www.youtube.com/embed/VWxukZc5k7s?rel=0&playsinline=1" title="Agent tracing with Failproof AI" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture; fullscreen" allowFullScreen style={{ position: "absolute", top: 0, left: 0, width: "100%", height: "100%", border: 0 }}></iframe>
 </div>
diff --git a/docs/start/quickstart.mdx b/docs/start/quickstart.mdx
index 7f5f893ad..88a080708 100644
--- a/docs/start/quickstart.mdx
+++ b/docs/start/quickstart.mdx
@@ -4,7 +4,9 @@ description: "Capture an agent session, find a failure, and start preventing it.
 icon: "zap"
 ---
 
-This quickstart gets one machine reporting sessions, runs an audit, and deploys a policy. Use the skill to set Failproof up, or follow the manual steps. You need Node.js 20.9 or later.
+This quickstart gets one machine reporting sessions, runs an audit, and deploys a policy. Use the skill to set Failproof up, or follow the manual steps.
+
+**Which path is yours?** If your agent runs in one of the 12 supported [harnesses](/reference/harnesses) — a coding CLI, or a gateway like Hermes or OpenClaw — follow the steps below; you need Node.js 20.9 or later. If your agent has no harness, instrument it with the [Python SDK](/reference/python-sdk) for tracing and audits, then rejoin at [Run your first failure check](/start/first-audit); enforcement on that path needs a hook in your runtime.
 
 <Tabs>
   <Tab title="Use the skill">
@@ -56,13 +58,16 @@ export FAILPROOFAI_KEY="<one-time-secret>"
     Open **Sessions** in Failproof AI and select an imported session.
   </Step>
   <Step title="Install agent hooks and policies">
-    This attaches Failproof AI to your agent CLI and installs the built-in policies. Use them to see local policy decisions and try enforcement before Failproof AI audits your sessions and writes policies for your agents.
+    This attaches Failproof AI to your harness and installs the 40 built-in policies. Use them to see local policy decisions and try enforcement before Failproof AI audits your sessions and writes policies for your agents.
 
-    Let the installer detect your agent CLI, or name one explicitly.
+    Let the installer detect your harness, or name one explicitly. Every one of the 12 is a valid `--cli` value — `claude`, `codex`, `copilot`, `cursor`, `opencode`, `pi`, `hermes`, `openclaw`, `factory`, `devin`, `antigravity`, `goose`.
 
     ```bash
-    failproofai policies --install --cli claude --scope user
+    failproofai policies --install --cli claude --scope user     # a coding CLI
+    failproofai policies --install --cli hermes --scope user     # a Slack/Telegram gateway
     ```
+
+    Blocking a tool call before it runs is verified on all 12. Turn-end gates are verified on 8 — see [enforcement capability](/reference/harnesses#enforcement-capability) for the per-harness matrix.
   </Step>
   <Step title="Audit the behavior">
     Follow [Run your first failure check](/start/first-audit). Use a concrete goal such as “find sessions where the agent retried a failing tool without changing its approach.”
diff --git a/package.json b/package.json
index 5ffdacaae..798a069a3 100644
--- a/package.json
+++ b/package.json
@@ -1,7 +1,7 @@
 {
   "name": "failproofai",
   "version": "1.0.2-beta.0",
-  "description": "The easiest way to manage policies that keep your AI agents reliable, on-task, and running autonomously — for Claude Code & the Agents SDK",
+  "description": "Observability and enforcement for AI agent harnesses. 40 built-in policies hooked into 12 of them — Claude Code, Codex, Cursor, Hermes, OpenClaw and more — blocking the tool call before it runs. Local dashboard included, no account needed.",
   "bin": {
     "failproofai": "./dist/cli.mjs",
     "failproofaid": "./bin/failproofaid-shim.mjs"
@@ -44,21 +44,26 @@
     "docs:audit": "bun scripts/docs-audit.ts"
   },
   "keywords": [
-    "claude",
-    "claude-code",
-    "claude-agents-sdk",
-    "anthropic",
-    "ai-agent",
-    "agent-reliability",
+    "agent-observability",
+    "llm-observability",
+    "agent-tracing",
     "agent-monitoring",
+    "agent-harness",
+    "guardrails",
+    "enforcement",
+    "policies",
+    "evals",
+    "ai-agent",
     "autonomous-agent",
+    "agent-reliability",
     "failure-prevention",
-    "developer-tools",
-    "devtools",
+    "claude-code",
+    "codex",
+    "hermes",
+    "openclaw",
     "cli",
     "local-first",
-    "hooks",
-    "policies"
+    "hooks"
   ],
   "author": "ExosphereHost Inc. <failproofai@exosphere.host>",
   "license": "SEE LICENSE IN LICENSE",
diff --git a/src/audit/report.ts b/src/audit/report.ts
index 31ede51b1..6ae87f783 100644
--- a/src/audit/report.ts
+++ b/src/audit/report.ts
@@ -263,16 +263,16 @@ export function formatMarkdown(result: AuditResult): string {
   out.push("## TL;DR");
   out.push("");
   if (totalHits === 0) {
-    out.push("Clean run — the AI coding agent didn't do anything `failproofai` catches in this window.");
+    out.push("Clean run — the agent didn't do anything `failproofai` catches in this window.");
   } else {
     out.push(
-      `Over ${result.transcripts.scanned} sessions, my AI coding agent did **${totalHits} things \`failproofai\` would have stopped**: ` +
-      `${totalProtected} were already blocked in real time by my current config; ` +
-      `**${totalSlipping} slipped through** (would've been caught if more policies were on).`,
+      `Over ${result.transcripts.scanned} sessions, my agent did **${totalHits} things \`failproofai\` caught**: ` +
+      `${totalProtected} were blocked in real time by my current config; ` +
+      `**${totalSlipping} slipped through** and were only caught by the audit.`,
     );
   }
   out.push("");
-  out.push("> [failproofai](https://github.com/FailproofAI/failproofai) is a hook-based policy engine for Claude Code, Codex, Copilot, Cursor, OpenCode, and Pi. The `audit` command replays past agent sessions through every builtin policy to surface patterns that were (or could've been) stopped.");
+  out.push("> [failproofai](https://github.com/FailproofAI/failproofai) is a hook-based policy engine for 12 agent harnesses — coding CLIs, chat gateways, and self-hosted assistants. The `audit` command replays past agent sessions through every builtin policy to surface patterns that were (or could've been) stopped.");
   out.push("");
 
   if (totalHits === 0) return out.join("\n");
diff --git a/src/hooks/configure-wizard.ts b/src/hooks/configure-wizard.ts
index 4fc90ad93..914a3e00a 100644
--- a/src/hooks/configure-wizard.ts
+++ b/src/hooks/configure-wizard.ts
@@ -6,7 +6,7 @@
  *   0. Daemon    — REQUIRED. Asks for sudo first, on a clean terminal.
  *   1. Where     — inferred from cwd, then confirmed (global / project / both)
  *   2. Policies  — multi-select of themed presets (combine any) or Everything
- *   3. Assistants— multi-select of agent CLIs (detected + install-ahead)
+ *   3. Assistants— multi-select of agent harnesses (detected + install-ahead)
  *   4. Connect   — paste an API key, or stay fully local
  *   5. Review    — shows exactly which files change, then Apply.
  *
diff --git a/src/hooks/manager.ts b/src/hooks/manager.ts
index 2c9dd4cef..baef48d8c 100644
--- a/src/hooks/manager.ts
+++ b/src/hooks/manager.ts
@@ -1,5 +1,5 @@
 /**
- * Install/remove/list failproofai hooks for one or more agent CLIs.
+ * Install/remove/list failproofai hooks for one or more agent harnesses.
  *
  * Per-CLI path resolution and settings I/O live in `./integrations` (one
  * `Integration` impl per CLI). This module orchestrates: validation, policy
diff --git a/src/hooks/types.ts b/src/hooks/types.ts
index b1021195b..fd2ff7f3a 100644
--- a/src/hooks/types.ts
+++ b/src/hooks/types.ts
@@ -1,5 +1,19 @@
 /**
- * Constants and interfaces for agent CLI hooks integrations (Claude Code, OpenAI Codex, GitHub Copilot, Cursor Agent, OpenCode, Pi, …).
+ * Constants and interfaces for agent harness hook integrations.
+ *
+ * A harness is whatever an agent runs inside. The twelve below are two classes:
+ * ten coding CLIs (Claude Code, Codex, Copilot, Cursor, OpenCode, Pi, Factory,
+ * Devin, Antigravity, Goose) and two chat/assistant gateways (Hermes, OpenClaw).
+ *
+ * An agent instrumented directly via the Python SDK is NOT one of these twelve
+ * and is not a third class here: the SDK reports events, it does not sit in the
+ * tool-call path, so it delivers observability without enforcement. Copy that
+ * lists it alongside the twelve overstates what it does.
+ *
+ * INTEGRATION_TYPES below is the authoritative list:
+ * __tests__/scripts/copy-counts.test.ts pins its length, and the policy, event
+ * and enforcement counts, against the harness and policy counts stated in
+ * README.md, package.json and docs/index.mdx.
  */
 
 export const HOOK_SCOPES = ["user", "project", "local"] as const;