agenticreplay 0.1.0 → 0.1.2

package/README.md

@@ -19,13 +19,6 @@ its `kiokuko mcp` child. Native Kiokuko DSH needs the companion recorder changes
 Provider-hidden reasoning and SDK changes before the capture boundary are not observable.
 
 [![License](https://img.shields.io/badge/code-Apache--2.0-blue)](LICENSE)
-[![Spec](https://img.shields.io/badge/trace%20spec-CC%20BY%204.0-blue)](spec/agenticreplay-trace-v0.md)
-[![Node](https://img.shields.io/badge/node-20%2B-brightgreen)](#install)
-[![Agents](https://img.shields.io/badge/agents-Claude%20Code%20%C2%B7%20Codex%20%C2%B7%20Agents%20SDK%20%C2%B7%20AI%20SDK%20%C2%B7%20any-black)](#which-agents)
-[![Good first issues](https://img.shields.io/badge/good%20first%20issues-12-orange)](docs/good-first-issues.md)
-[![Mentioned in Awesome Claude Code](https://awesome.re/mentioned-badge.svg)](https://github.com/hesreallyhim/awesome-claude-code)
-
-![Recording a Claude Code run, replaying it offline, then forking it onto two models](docs/demo-cli.gif)
 
 <sup>Real output from one session — a Claude Code run recorded, replayed against the recording, then
 forked at checkpoint 4 onto two models and graded by `npx tsc --noEmit`. Nothing here is mocked up.</sup>
@@ -791,20 +784,28 @@ agenticreplay doctor                       # checks node, git, and which agents
 Node 24 or newer. No native dependencies, so there is nothing to compile and nothing fetched at
 install time beyond the tarballs themselves.
 
-Every release from 0.1.1 onward is published by the tagged workflow in
-[`RELEASING.md`](RELEASING.md), with a provenance attestation naming the commit and the run that
-built it — npm shows it on the package page, and `npm audit signatures` checks it.
+The CLI package bundles its runtime dependencies, including the internal workspace libraries.
 
 **From source**, to work on it or to run an unreleased commit:
 
 ```console
 git clone https://github.com/askdkc/AgenticReplay && cd AgenticReplay
 npm ci && npm run build
-npm install -g ./packages/cli     # puts `agenticreplay` (and `agenticreplay`) on PATH
+npm install -g .                 # puts `agenticreplay` on PATH
+```
+
+To release from the repository root:
+
+```console
+npm ci
+npm publish
 ```
 
-`npm install -g .` from the repository root installs nothing: the root is a workspace with no
-binary of its own, and `agenticreplay` lives in `packages/cli`.
+The `prepack` lifecycle builds the CLI and validates the workspace dependency versions. npm packs
+the binary, quickstart assets, and runtime dependencies before uploading them. To create a local
+archive without uploading, run `npm pack`; it writes `agenticreplay-<version>.tgz`. Adding
+`--dry-run` only previews the archive. `node scripts/installed-smoke.mjs` verifies one actual root
+archive in a clean offline installation.
 
 **Node 24+ to run it** (the CLI's `engines` says `>=24.0.0`). Development and CI use Node 24 or
 26. No account, no signup, no API key changes.

package/bin/agenticreplay.js

@@ -0,0 +1,2 @@
+#!/usr/bin/env node
+import '../packages/cli/dist/cli.js';

package/node_modules/@agenticreplay/adapters/dist/claude-code.d.ts

@@ -0,0 +1,37 @@
+import type { Adapter, SessionSupport } from '@agenticreplay/plugin-api';
+/**
+ * Claude Code names a project directory after the working directory, with every character that is
+ * not a letter, a digit or a hyphen replaced by one.
+ *
+ * Checked against a real install rather than inferred: a directory called `slug_test.dir v2`
+ * becomes `slug-test-dir-v2`, so underscores, dots and spaces all fold the same way the path
+ * separators do. An earlier version replaced only `\`, `/` and `:`, which was right for every
+ * path it happened to be tested against and wrong for the first one with an underscore in it —
+ * the computed name missed the real directory by one character, and the session capture then
+ * found nothing and said nothing. Hence {@link claudeSession}'s fallback: a guess about someone
+ * else's naming should not be the only thing holding the capture up.
+ */
+export declare function claudeProjectSlug(cwd: string): string;
+/**
+ * Claude Code's transcript.
+ *
+ * `user` entries carry what the person typed, which is the half of a run agenticreplay cannot see from the
+ * wire. Entries the harness generates for itself — tool results, the attachments it splices in —
+ * arrive with the same `user` role, so they are filtered out: a replay driven by a tool result
+ * would ask the model something the recording never asked.
+ */
+export declare const claudeSession: SessionSupport;
+/**
+ * Claude Code reads `ANTHROPIC_BASE_URL` for the API origin and appends its own `/v1/...` paths,
+ * so the proxy url goes in bare.
+ *
+ * No placeholder key. The placeholder exists for SDK clients that refuse to construct without one,
+ * and Claude Code is not one: it carries its own credential store, and most installations sign in
+ * to claude.ai rather than setting `ANTHROPIC_API_KEY` at all. Inventing the variable for those
+ * costs the recording twice over — Claude Code stops on an interactive "detected a custom API key"
+ * prompt before it will start, and answering yes switches it off the subscription credential it
+ * was going to authenticate with and onto a key that is not real. Passing the variable through
+ * when it is genuinely set keeps the API-key path exactly as it was.
+ */
+export declare const claudeCodeAdapter: Adapter;
+//# sourceMappingURL=claude-code.d.ts.map

package/node_modules/@agenticreplay/adapters/dist/claude-code.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-code.d.ts","sourceRoot":"","sources":["../src/claude-code.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,OAAO,EAAyB,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAKhG;;;;;;;;;;;GAWG;AACH,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAErD;AAED;;;;;;;GAOG;AACH,eAAO,MAAM,aAAa,EAAE,cA8B3B,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,eAAO,MAAM,iBAAiB,EAAE,OAiC/B,CAAC"}

package/node_modules/@agenticreplay/adapters/dist/claude-code.js

@@ -0,0 +1,103 @@
+import { existsSync } from 'node:fs';
+import { join } from 'node:path';
+import { detectAgent, homeDir } from './detect.js';
+import { passThrough, proxyBase, readEnv } from './env.js';
+import { contentText, parseJsonl } from './session.js';
+/**
+ * Claude Code names a project directory after the working directory, with every character that is
+ * not a letter, a digit or a hyphen replaced by one.
+ *
+ * Checked against a real install rather than inferred: a directory called `slug_test.dir v2`
+ * becomes `slug-test-dir-v2`, so underscores, dots and spaces all fold the same way the path
+ * separators do. An earlier version replaced only `\`, `/` and `:`, which was right for every
+ * path it happened to be tested against and wrong for the first one with an underscore in it —
+ * the computed name missed the real directory by one character, and the session capture then
+ * found nothing and said nothing. Hence {@link claudeSession}'s fallback: a guess about someone
+ * else's naming should not be the only thing holding the capture up.
+ */
+export function claudeProjectSlug(cwd) {
+    return cwd.replace(/[^A-Za-z0-9-]/g, '-');
+}
+/**
+ * Claude Code's transcript.
+ *
+ * `user` entries carry what the person typed, which is the half of a run agenticreplay cannot see from the
+ * wire. Entries the harness generates for itself — tool results, the attachments it splices in —
+ * arrive with the same `user` role, so they are filtered out: a replay driven by a tool result
+ * would ask the model something the recording never asked.
+ */
+export const claudeSession = {
+    dir(cwd, env) {
+        const home = readEnv(env, 'CLAUDE_CONFIG_DIR') ?? join(homeDir(), '.claude');
+        const projects = join(home, 'projects');
+        const named = join(projects, claudeProjectSlug(cwd));
+        // The slug is a guess about someone else's naming, and a wrong guess loses the capture in
+        // silence. When the directory it names is not there, watch the whole projects tree instead:
+        // the file that changed during this run is still this run's, whatever it ended up called.
+        return existsSync(named) ? named : projects;
+    },
+    parse(bytes) {
+        const prompts = [];
+        let id;
+        for (const entry of parseJsonl(bytes)) {
+            if (typeof entry['sessionId'] === 'string')
+                id ??= entry['sessionId'];
+            if (entry['type'] !== 'user')
+                continue;
+            const message = entry['message'];
+            if (message === null || typeof message !== 'object')
+                continue;
+            const text = contentText(message.content).trim();
+            // A turn the harness synthesised, not one a person typed.
+            if (text === '' || text.startsWith('<'))
+                continue;
+            prompts.push(text);
+        }
+        return { id, prompts };
+    },
+    resumeArgs(id) {
+        return ['--resume', id];
+    },
+};
+/**
+ * Claude Code reads `ANTHROPIC_BASE_URL` for the API origin and appends its own `/v1/...` paths,
+ * so the proxy url goes in bare.
+ *
+ * No placeholder key. The placeholder exists for SDK clients that refuse to construct without one,
+ * and Claude Code is not one: it carries its own credential store, and most installations sign in
+ * to claude.ai rather than setting `ANTHROPIC_API_KEY` at all. Inventing the variable for those
+ * costs the recording twice over — Claude Code stops on an interactive "detected a custom API key"
+ * prompt before it will start, and answering yes switches it off the subscription credential it
+ * was going to authenticate with and onto a key that is not real. Passing the variable through
+ * when it is genuinely set keeps the API-key path exactly as it was.
+ */
+export const claudeCodeAdapter = {
+    id: 'claude-code',
+    // What people type. The binary is `claude`, and so is every example in the docs.
+    aliases: ['claude'],
+    harnessVersions: '>=1.0.0',
+    async detect(_cwd) {
+        return detectAgent(['claude'], ['.claude']);
+    },
+    async prepare(ctx) {
+        const env = { ANTHROPIC_BASE_URL: proxyBase(ctx.proxyUrl) };
+        passThrough(env, ctx.env, 'ANTHROPIC_API_KEY');
+        passThrough(env, ctx.env, 'ANTHROPIC_AUTH_TOKEN');
+        return { command: 'claude', args: [...ctx.userArgs], env };
+    },
+    session: claudeSession,
+    /**
+     * `--strict-mcp-config` alongside it, so the run uses the instrumented copy *instead of* the
+     * user's own servers rather than as well as them. Without it a project `.mcp.json` is still
+     * loaded, the same server is registered twice, and half its calls go round the shim.
+     */
+    mcpConfigArgs(path) {
+        return ['--mcp-config', path, '--strict-mcp-config'];
+    },
+    /** `-p` is how Claude Code takes a prompt without a terminal, which a replay never has. */
+    driveArgs(prompts, recorded) {
+        const first = prompts[0];
+        return first === undefined ? undefined : [...recorded, '-p', first];
+    },
+};
+//# sourceMappingURL=claude-code.js.map

package/node_modules/@agenticreplay/adapters/dist/claude-code.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-code.js","sourceRoot":"","sources":["../src/claude-code.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,OAAO,EAAE,WAAW,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AACnD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,UAAU,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAEvD;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAAC,GAAW;IAC3C,OAAO,GAAG,CAAC,OAAO,CAAC,gBAAgB,EAAE,GAAG,CAAC,CAAC;AAC5C,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,aAAa,GAAmB;IAC3C,GAAG,CAAC,GAAG,EAAE,GAAG;QACV,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,EAAE,mBAAmB,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,SAAS,CAAC,CAAC;QAC7E,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;QACxC,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,EAAE,iBAAiB,CAAC,GAAG,CAAC,CAAC,CAAC;QACrD,0FAA0F;QAC1F,4FAA4F;QAC5F,0FAA0F;QAC1F,OAAO,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;IAC9C,CAAC;IAED,KAAK,CAAC,KAAK;QACT,MAAM,OAAO,GAAa,EAAE,CAAC;QAC7B,IAAI,EAAsB,CAAC;QAC3B,KAAK,MAAM,KAAK,IAAI,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;YACtC,IAAI,OAAO,KAAK,CAAC,WAAW,CAAC,KAAK,QAAQ;gBAAE,EAAE,KAAK,KAAK,CAAC,WAAW,CAAC,CAAC;YACtE,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,MAAM;gBAAE,SAAS;YACvC,MAAM,OAAO,GAAG,KAAK,CAAC,SAAS,CAAC,CAAC;YACjC,IAAI,OAAO,KAAK,IAAI,IAAI,OAAO,OAAO,KAAK,QAAQ;gBAAE,SAAS;YAC9D,MAAM,IAAI,GAAG,WAAW,CAAE,OAAiC,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;YAC5E,0DAA0D;YAC1D,IAAI,IAAI,KAAK,EAAE,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC;gBAAE,SAAS;YAClD,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACrB,CAAC;QACD,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC;IACzB,CAAC;IAED,UAAU,CAAC,EAAE;QACX,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IAC1B,CAAC;CACF,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAY;IACxC,EAAE,EAAE,aAAa;IACjB,iFAAiF;IACjF,OAAO,EAAE,CAAC,QAAQ,CAAC;IACnB,eAAe,EAAE,SAAS;IAE1B,KAAK,CAAC,MAAM,CAAC,IAAY;QACvB,OAAO,WAAW,CAAC,CAAC,QAAQ,CAAC,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;IAC9C,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,GAAkB;QAC9B,MAAM,GAAG,GAA2B,EAAE,kBAAkB,EAAE,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpF,WAAW,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,mBAAmB,CAAC,CAAC;QAC/C,WAAW,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,sBAAsB,CAAC,CAAC;QAClD,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,GAAG,GAAG,CAAC,QAAQ,CAAC,EAAE,GAAG,EAAE,CAAC;IAC7D,CAAC;IAED,OAAO,EAAE,aAAa;IAEtB;;;;OAIG;IACH,aAAa,CAAC,IAAI;QAChB,OAAO,CAAC,cAAc,EAAE,IAAI,EAAE,qBAAqB,CAAC,CAAC;IACvD,CAAC;IAED,2FAA2F;IAC3F,SAAS,CAAC,OAAO,EAAE,QAAQ;QACzB,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QACzB,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,GAAG,QAAQ,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC;IACtE,CAAC;CACF,CAAC"}

package/node_modules/@agenticreplay/adapters/dist/codex.d.ts

@@ -0,0 +1,36 @@
+import type { Adapter, SessionSupport } from '@agenticreplay/plugin-api';
+/**
+ * Which provider `config.toml` selects, defaulting to the built-in `openai`.
+ *
+ * A hand-rolled read of one key rather than a TOML parser: this is a top-level assignment in a
+ * file Codex writes itself, and a dependency for it would be the only one in this package.
+ * Anything unparseable falls back to the default, because a malformed config must degrade the
+ * capture rather than stop the run.
+ */
+export declare function selectedProvider(configToml: string): string;
+/**
+ * Codex's rollout.
+ *
+ * The user's turns arrive as `response_item` messages whose content is `input_text`. The harness
+ * puts its own preamble through the same channel — skills, environment context, multi-agent
+ * instructions — and those are wrapped in a tag, which is what separates them from a typed turn.
+ */
+export declare const codexSession: SessionSupport;
+/**
+ * Codex CLI.
+ *
+ * `OPENAI_BASE_URL` routes the built-in provider. Recent Codex releases reject overrides of
+ * reserved built-in provider IDs. Custom providers still require a scoped config override: Codex resolves its origin from `model_providers.<name>.base_url` in `config.toml`, so an
+ * installation pointed at a gateway — which is the reason most people have a `config.toml` at all
+ * — ignores the variable completely and talks straight to its own endpoint. The recording then
+ * ends with `capture.empty` and nothing to replay, having cost a full run of real tokens.
+ *
+ * `-c` is Codex's own flag for overriding exactly one config value for one invocation, so the
+ * override is scoped to the run and the user's file is never touched. The provider name has to be
+ * read from that file, because the key to override is named after whichever provider is selected.
+ *
+ * No placeholder key: Codex carries its own credential store in `auth.json`, and inventing
+ * `OPENAI_API_KEY` overrides it with something that is not real.
+ */
+export declare const codexAdapter: Adapter;
+//# sourceMappingURL=codex.d.ts.map

package/node_modules/@agenticreplay/adapters/dist/codex.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"codex.d.ts","sourceRoot":"","sources":["../src/codex.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,OAAO,EAAyB,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAUhG;;;;;;;GAOG;AACH,wBAAgB,gBAAgB,CAAC,UAAU,EAAE,MAAM,GAAG,MAAM,CAU3D;AAED;;;;;;GAMG;AACH,eAAO,MAAM,YAAY,EAAE,cAkC1B,CAAC;AAEF;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,YAAY,EAAE,OA2C1B,CAAC"}

package/node_modules/@agenticreplay/adapters/dist/codex.js

@@ -0,0 +1,130 @@
+import { readFile } from 'node:fs/promises';
+import { join } from 'node:path';
+import { detectAgent, homeDir } from './detect.js';
+import { passThrough, proxyBase, readEnv } from './env.js';
+import { contentText, parseJsonl } from './session.js';
+/** Where the Codex CLI keeps `config.toml`, honouring its own override. */
+function codexHome(env) {
+    return readEnv(env, 'CODEX_HOME') ?? join(homeDir(), '.codex');
+}
+/**
+ * Which provider `config.toml` selects, defaulting to the built-in `openai`.
+ *
+ * A hand-rolled read of one key rather than a TOML parser: this is a top-level assignment in a
+ * file Codex writes itself, and a dependency for it would be the only one in this package.
+ * Anything unparseable falls back to the default, because a malformed config must degrade the
+ * capture rather than stop the run.
+ */
+export function selectedProvider(configToml) {
+    for (const line of configToml.split(/\r?\n/)) {
+        const text = line.trim();
+        // Stop at the first table header: `model_provider` is a top-level key, and a same-named key
+        // inside `[model_providers.x]` is a different setting entirely.
+        if (text.startsWith('['))
+            break;
+        const match = /^model_provider\s*=\s*["']([^"']+)["']/.exec(text);
+        if (match?.[1])
+            return match[1];
+    }
+    return 'openai';
+}
+/**
+ * Codex's rollout.
+ *
+ * The user's turns arrive as `response_item` messages whose content is `input_text`. The harness
+ * puts its own preamble through the same channel — skills, environment context, multi-agent
+ * instructions — and those are wrapped in a tag, which is what separates them from a typed turn.
+ */
+export const codexSession = {
+    dir(_cwd, env) {
+        return join(codexHome(env), 'sessions');
+    },
+    parse(bytes) {
+        const prompts = [];
+        let id;
+        for (const entry of parseJsonl(bytes)) {
+            const payload = entry['payload'];
+            if (payload === null || typeof payload !== 'object')
+                continue;
+            const p = payload;
+            if (entry['type'] === 'session_meta' && typeof p['id'] === 'string')
+                id ??= p['id'];
+            if (entry['type'] !== 'response_item' || p['type'] !== 'message')
+                continue;
+            if (p['role'] !== undefined && p['role'] !== 'user')
+                continue;
+            const content = p['content'];
+            if (!Array.isArray(content))
+                continue;
+            const text = content
+                .filter((b) => b !== null && typeof b === 'object' && b.type === 'input_text')
+                .map((b) => String(b.text ?? ''))
+                .join('')
+                .trim();
+            if (text === '' || text.startsWith('<'))
+                continue;
+            prompts.push(text);
+        }
+        return { id, prompts };
+    },
+    resumeArgs(id) {
+        return ['exec', 'resume', id];
+    },
+};
+/**
+ * Codex CLI.
+ *
+ * `OPENAI_BASE_URL` routes the built-in provider. Recent Codex releases reject overrides of
+ * reserved built-in provider IDs. Custom providers still require a scoped config override: Codex resolves its origin from `model_providers.<name>.base_url` in `config.toml`, so an
+ * installation pointed at a gateway — which is the reason most people have a `config.toml` at all
+ * — ignores the variable completely and talks straight to its own endpoint. The recording then
+ * ends with `capture.empty` and nothing to replay, having cost a full run of real tokens.
+ *
+ * `-c` is Codex's own flag for overriding exactly one config value for one invocation, so the
+ * override is scoped to the run and the user's file is never touched. The provider name has to be
+ * read from that file, because the key to override is named after whichever provider is selected.
+ *
+ * No placeholder key: Codex carries its own credential store in `auth.json`, and inventing
+ * `OPENAI_API_KEY` overrides it with something that is not real.
+ */
+export const codexAdapter = {
+    id: 'codex',
+    async detect(_cwd) {
+        return detectAgent(['codex'], ['.codex']);
+    },
+    async prepare(ctx) {
+        const base = proxyBase(ctx.proxyUrl, 'v1');
+        const env = { OPENAI_BASE_URL: base };
+        passThrough(env, ctx.env, 'OPENAI_API_KEY');
+        let provider = 'openai';
+        try {
+            provider = selectedProvider(await readFile(join(codexHome(ctx.env), 'config.toml'), 'utf8'));
+        }
+        catch {
+            // No config, or unreadable: the built-in provider is the right guess, and `OPENAI_BASE_URL`
+            // covers it as well. Capture must never depend on a file existing.
+        }
+        // Ahead of the user's own arguments so that anything they pass explicitly still wins.
+        const args = provider === 'openai'
+            ? ['-c', `openai_base_url=${base}`, ...ctx.userArgs]
+            : ['-c', `model_providers.${provider}.base_url=${base}`, ...ctx.userArgs];
+        return { command: 'codex', args, env };
+    },
+    session: codexSession,
+    /**
+     * `exec` is Codex's non-interactive form and takes the prompt as its positional argument.
+     *
+     * A run recorded as `codex exec` with the prompt on stdin already carries the subcommand and its
+     * flags, so the prompt is appended to what was recorded; a run recorded bare needs the whole
+     * form. `--skip-git-repo-check` because a replay runs in a scratch directory that is not one.
+     */
+    driveArgs(prompts, recorded) {
+        const first = prompts[0];
+        if (first === undefined)
+            return undefined;
+        if (recorded.includes('exec'))
+            return [...recorded, first];
+        return ['exec', '--skip-git-repo-check', first];
+    },
+};
+//# sourceMappingURL=codex.js.map

package/node_modules/@agenticreplay/adapters/dist/codex.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"codex.js","sourceRoot":"","sources":["../src/codex.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,OAAO,EAAE,WAAW,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AACnD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,UAAU,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAEvD,2EAA2E;AAC3E,SAAS,SAAS,CAAC,GAAuC;IACxD,OAAO,OAAO,CAAC,GAAG,EAAE,YAAY,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,CAAC,CAAC;AACjE,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,gBAAgB,CAAC,UAAkB;IACjD,KAAK,MAAM,IAAI,IAAI,UAAU,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;QAC7C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,4FAA4F;QAC5F,gEAAgE;QAChE,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC;YAAE,MAAM;QAChC,MAAM,KAAK,GAAG,wCAAwC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAClE,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC;YAAE,OAAO,KAAK,CAAC,CAAC,CAAC,CAAC;IAClC,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,YAAY,GAAmB;IAC1C,GAAG,CAAC,IAAI,EAAE,GAAG;QACX,OAAO,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,UAAU,CAAC,CAAC;IAC1C,CAAC;IAED,KAAK,CAAC,KAAK;QACT,MAAM,OAAO,GAAa,EAAE,CAAC;QAC7B,IAAI,EAAsB,CAAC;QAC3B,KAAK,MAAM,KAAK,IAAI,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;YACtC,MAAM,OAAO,GAAG,KAAK,CAAC,SAAS,CAAC,CAAC;YACjC,IAAI,OAAO,KAAK,IAAI,IAAI,OAAO,OAAO,KAAK,QAAQ;gBAAE,SAAS;YAC9D,MAAM,CAAC,GAAG,OAAkC,CAAC;YAC7C,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,cAAc,IAAI,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,QAAQ;gBAAE,EAAE,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC;YACpF,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,eAAe,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,SAAS;gBAAE,SAAS;YAC3E,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,MAAM;gBAAE,SAAS;YAC9D,MAAM,OAAO,GAAG,CAAC,CAAC,SAAS,CAAC,CAAC;YAC7B,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;gBAAE,SAAS;YACtC,MAAM,IAAI,GAAG,OAAO;iBACjB,MAAM,CACL,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,KAAK,IAAI,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAK,CAAuB,CAAC,IAAI,KAAK,YAAY,CACxF;iBACA,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAE,CAAwB,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;iBACxD,IAAI,CAAC,EAAE,CAAC;iBACR,IAAI,EAAE,CAAC;YACV,IAAI,IAAI,KAAK,EAAE,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC;gBAAE,SAAS;YAClD,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACrB,CAAC;QACD,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC;IACzB,CAAC;IAED,UAAU,CAAC,EAAE;QACX,OAAO,CAAC,MAAM,EAAE,QAAQ,EAAE,EAAE,CAAC,CAAC;IAChC,CAAC;CACF,CAAC;AAEF;;;;;;;;;;;;;;;GAeG;AACH,MAAM,CAAC,MAAM,YAAY,GAAY;IACnC,EAAE,EAAE,OAAO;IAEX,KAAK,CAAC,MAAM,CAAC,IAAY;QACvB,OAAO,WAAW,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC,CAAC;IAC5C,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,GAAkB;QAC9B,MAAM,IAAI,GAAG,SAAS,CAAC,GAAG,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QAC3C,MAAM,GAAG,GAA2B,EAAE,eAAe,EAAE,IAAI,EAAE,CAAC;QAC9D,WAAW,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,gBAAgB,CAAC,CAAC;QAE5C,IAAI,QAAQ,GAAG,QAAQ,CAAC;QACxB,IAAI,CAAC;YACH,QAAQ,GAAG,gBAAgB,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,aAAa,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;QAC/F,CAAC;QAAC,MAAM,CAAC;YACP,4FAA4F;YAC5F,mEAAmE;QACrE,CAAC;QAED,sFAAsF;QACtF,MAAM,IAAI,GACR,QAAQ,KAAK,QAAQ;YACnB,CAAC,CAAC,CAAC,IAAI,EAAE,mBAAmB,IAAI,EAAE,EAAE,GAAG,GAAG,CAAC,QAAQ,CAAC;YACpD,CAAC,CAAC,CAAC,IAAI,EAAE,mBAAmB,QAAQ,aAAa,IAAI,EAAE,EAAE,GAAG,GAAG,CAAC,QAAQ,CAAC,CAAC;QAC9E,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,CAAC;IACzC,CAAC;IAED,OAAO,EAAE,YAAY;IAErB;;;;;;OAMG;IACH,SAAS,CAAC,OAAO,EAAE,QAAQ;QACzB,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QACzB,IAAI,KAAK,KAAK,SAAS;YAAE,OAAO,SAAS,CAAC;QAC1C,IAAI,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC;YAAE,OAAO,CAAC,GAAG,QAAQ,EAAE,KAAK,CAAC,CAAC;QAC3D,OAAO,CAAC,MAAM,EAAE,uBAAuB,EAAE,KAAK,CAAC,CAAC;IAClD,CAAC;CACF,CAAC"}

package/node_modules/@agenticreplay/adapters/dist/contract.d.ts

@@ -0,0 +1,65 @@
+import type { Adapter, RecordContext } from '@agenticreplay/plugin-api';
+/**
+ * The adapter contract: the behaviour every adapter must have for capture to work at all.
+ *
+ * Adapters break silently. A harness renames the variable it reads for its API origin, the adapter
+ * keeps setting the old one, and every run still *looks* fine — the agent answers, the exit code is
+ * zero, and the trace is empty. Each check below exists because breaking it produces exactly that
+ * shape of failure: nothing to see until a user files a confusing bug.
+ */
+/** Every check, in the order they run. A failing check never stops the others. */
+export declare const CONTRACT_CHECKS: readonly ["id-format", "detect-resolves", "prepare-shape", "redirects-model-traffic", "no-invented-keys", "no-ctx-mutation", "deterministic", "no-foreign-paths", "harness-versions", "artifacts-resettable"];
+export type ContractCheck = (typeof CONTRACT_CHECKS)[number];
+export interface ContractFailure {
+    check: ContractCheck;
+    detail: string;
+}
+export interface ContractResult {
+    adapter: string;
+    passed: string[];
+    failed: ContractFailure[];
+    ok: boolean;
+}
+export interface ContractOptions {
+    /**
+     * Overrides for the context the checks prepare against — `userArgs` for an adapter that takes
+     * its command from argv, or a `proxyUrl` matching a running proxy. `cwd` and `runDir` default to
+     * fresh temp directories that are removed afterwards; `env` is owned by the credential checks,
+     * which vary it deliberately.
+     */
+    ctx?: Partial<RecordContext>;
+}
+/** At least one of these must be redirected, or the proxy sees nothing at all. */
+export declare const MODEL_BASE_URL_VARS: readonly ["ANTHROPIC_BASE_URL", "OPENAI_BASE_URL", "OPENAI_API_BASE"];
+/**
+ * Runs one adapter through the contract. Never throws: a runner that dies on a broken adapter is
+ * the same class of bug it is trying to catch, and callers batch this over a whole registry.
+ */
+export declare function checkAdapterContract(adapter: Adapter, opts?: ContractOptions): Promise<ContractResult>;
+/** One line for a pass, one line per failure otherwise — the shape a CI log wants. */
+export declare function formatContractResult(result: ContractResult): string;
+/**
+ * Whether `path` is `base` or sits under it.
+ *
+ * Delegated to the platform's own path rules rather than compared as strings. The prefix used to be
+ * built with a forward slash only, so on Windows — where `path.join` spells the whole path with
+ * backslashes — a file genuinely inside `runDir` never matched, and `no-foreign-paths` called the
+ * run's own scratch file foreign. That failed the contract for every adapter that puts a runDir
+ * path into the environment: grok, openclaw and node. On POSIX the comparison was true either way,
+ * so CI never saw it.
+ *
+ * Normalising the separators by hand would have swapped one wrong answer for another: a backslash
+ * is a legal character in a POSIX filename, so `/tmp/run` + a file called `run\evil` would have
+ * read as contained when it is not — a leak check answering yes to the thing it exists to catch.
+ *
+ * `impl` is a parameter so both flavours are reachable from a test on either OS. The Windows
+ * behaviour is the whole reason this function was touched, and a test that can only run on Windows
+ * is how the original slipped through.
+ */
+export declare function isInside(path: string, base: string, impl?: PathImpl): boolean;
+/** The slice of `node:path` {@link isInside} needs, so `path.win32` and `path.posix` both fit. */
+export interface PathImpl {
+    relative(from: string, to: string): string;
+    isAbsolute(p: string): boolean;
+}
+//# sourceMappingURL=contract.d.ts.map

package/node_modules/@agenticreplay/adapters/dist/contract.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"contract.d.ts","sourceRoot":"","sources":["../src/contract.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,OAAO,EAAU,aAAa,EAAE,MAAM,2BAA2B,CAAC;AAGhF;;;;;;;GAOG;AAEH,kFAAkF;AAClF,eAAO,MAAM,eAAe,+MAWlB,CAAC;AAEX,MAAM,MAAM,aAAa,GAAG,CAAC,OAAO,eAAe,CAAC,CAAC,MAAM,CAAC,CAAC;AAE7D,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,aAAa,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,MAAM,EAAE,eAAe,EAAE,CAAC;IAC1B,EAAE,EAAE,OAAO,CAAC;CACb;AAED,MAAM,WAAW,eAAe;IAC9B;;;;;OAKG;IACH,GAAG,CAAC,EAAE,OAAO,CAAC,aAAa,CAAC,CAAC;CAC9B;AAED,kFAAkF;AAClF,eAAO,MAAM,mBAAmB,uEAItB,CAAC;AA4BX;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,OAAO,EAChB,IAAI,GAAE,eAAoB,GACzB,OAAO,CAAC,cAAc,CAAC,CAsCzB;AAED,sFAAsF;AACtF,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CAKnE;AAqVD;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,GAAE,QAAmB,GAAG,OAAO,CAIvF;AAED,kGAAkG;AAClG,MAAM,WAAW,QAAQ;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,GAAG,MAAM,CAAC;IAC3C,UAAU,CAAC,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC;CAChC"}