@atrib/summarize 0.2.0

package/LICENSE

@@ -0,0 +1,190 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship made available under
+      the License, as indicated by a copyright notice that is included in
+      or attached to the work (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean, as submitted to the Licensor for inclusion
+      in the Work by the copyright owner or by an individual or Legal Entity
+      authorized to submit on behalf of the copyright owner. For the purposes
+      of this definition, "submitted" means any form of electronic, verbal,
+      or written communication sent to the Licensor or its representatives,
+      including but not limited to communication on electronic mailing lists,
+      source code control systems, and issue tracking systems that are managed
+      by, or on behalf of, the Licensor for the purpose of discussing and
+      improving the Work, but excluding communication that is conspicuously
+      marked or otherwise designated in writing by the copyright owner as
+      "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any Legal Entity on behalf of
+      whom a Contribution has been received by the Licensor and subsequently
+      incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a cross-claim
+      or counterclaim in a lawsuit) alleging that the Work or a Contribution
+      incorporated within the Work constitutes direct or contributory patent
+      infringement, then any patent licenses granted to You under this License
+      for that Work shall terminate as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or Derivative Works
+          a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, You must include a readable copy of the attribution
+          notices contained within such NOTICE file, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own license statement for Your modifications and
+      may provide additional grant of rights to use, modify, or distribute
+      those modifications in accordance with this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or reproducing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or exemplary damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or all other
+      commercial damages or losses), even if such Contributor has been
+      advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format in question. Also, an optional
+      "Statement of Purpose" appears after the copyright notice.
+
+   Copyright 2025-2026 Atrib contributors
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+   implied. See the License for the specific language governing
+   permissions and limitations under the License.

package/README.md

@@ -0,0 +1,78 @@
+# @atrib/atrib-summarize
+
+MCP server exposing the `summarize` tool — synthesizes a narrative across N records using an OpenAI-compatible LLM.
+
+Closes the consumer-side cognitive-loop primitive companion to `atrib-trace`: trace returns the causal chain (structural); summarize returns the synthesized meaning across the chain (semantic). Both read the same local mirror including the optional `_local` sidecar.
+
+## Tool
+
+```
+mcp__atrib-summarize__summarize({
+  context_id?: string,        // 32-hex trace; summarize all records in it
+  record_hashes?: string[],   // explicit list; unioned with context_id
+  focus?: string,             // optional steering for the synthesis
+  max_records?: number,       // cap on records fed to LLM (default 50, max 200)
+  model?: string              // override model from env
+})
+→ {
+  narrative: string | null,
+  cited_record_hashes: string[],
+  records_summarized: number,
+  records_skipped: number,           // beyond max_records
+  records_with_sidecar: number,      // had semantic content
+  records_without_sidecar: number,   // legacy bare records, impoverished input
+  model_used: string | null,
+  warnings: string[]
+}
+```
+
+## LLM provider
+
+OpenAI-compatible HTTP. Defaults to NVIDIA NIM with `qwen/qwen3.5-397b-a17b`. Override via env or per-call `model` input:
+
+| Env var | Default |
+|---|---|
+| `ATRIB_SUMMARIZE_API_KEY` | (fallback to `NVIDIA_API_KEY` then `NVIDIA_NIM_API_KEY`) |
+| `ATRIB_SUMMARIZE_BASE_URL` | `https://integrate.api.nvidia.com/v1` |
+| `ATRIB_SUMMARIZE_MODEL` | `qwen/qwen3.5-397b-a17b` |
+| `ATRIB_SUMMARIZE_MAX_TOKENS` | `4000` |
+| `ATRIB_SUMMARIZE_TEMPERATURE` | `0.3` |
+| `ATRIB_SUMMARIZE_TIMEOUT_MS` | `120000` |
+
+Without an API key, the tool returns a warnings-only response per the §5.8 graceful-degradation contract.
+
+## Reads
+
+Same as `@atrib/atrib-trace`: every `*.jsonl` mirror under `~/.atrib/records/` (override via `ATRIB_RECORDS_DIR`). Tolerates both legacy bare-record and current envelope shapes.
+
+When a record lacks a `_local` sidecar (legacy entry), the prompt includes a marker telling the LLM the input is impoverished — only event_type + cryptographic metadata is available — so the synthesis can be honest about gaps. The output reports `records_with_sidecar` and `records_without_sidecar` counts so callers know how rich the input was.
+
+## Behaviors
+
+- **Selection**: `record_hashes` ∪ records-with-matching-`context_id`, deduplicated.
+- **Capping**: chronological-ascending sort, then take `max_records`. Skipped count surfaced.
+- **Honest input flagging**: a per-record line in the prompt marks records lacking semantic content; the system prompt instructs the LLM not to invent semantics.
+- **Network access only on LLM call**: storage reads are local. Tests use the same FORBIDDEN_HOSTS guard as the rest of the workspace to prevent fixture leakage.
+- **No retry on LLM failure**: surfaces the error in `warnings` and returns null narrative. Caller decides whether to retry with smaller `max_records` or different model.
+
+## Wire-up
+
+Add to your MCP host config:
+
+```json
+{
+  "atrib-summarize": {
+    "command": "node",
+    "args": ["/path/to/atrib-summarize/dist/main.js"],
+    "env": {
+      "NVIDIA_API_KEY": "..."
+    }
+  }
+}
+```
+
+## Status
+
+Initial scaffold (v0.1.0). 6 unit tests covering record selection (by hash, by context, unioned, missing-skip) + degradation paths (no inputs, no API key). Integration test against a real LLM is gated behind `ATRIB_SUMMARIZE_API_KEY` and not run in CI.
+
+The companion `atrib-trace` is the structural primitive; together they close the consumer side of the cognitive loop.

package/dist/index.d.ts

@@ -0,0 +1,58 @@
+/**
+ * atrib-summarize MCP server: registers the `summarize` tool that reads
+ * a set of records (by context_id, by record_hashes, or both) and asks
+ * an OpenAI-compatible LLM to synthesize a narrative across them.
+ *
+ * Closes the consumer-side cognitive-loop primitive companion to trace:
+ * trace returns the causal chain (structural); summarize returns the
+ * synthesized meaning across that chain (semantic). Both read the same
+ * local mirror including the optional _local sidecar.
+ *
+ * LLM access via OpenAI-compatible HTTP. Defaults to NVIDIA NIM with the
+ * Qwen 397B model; override via env vars or per-call model override.
+ * Without an API key, the tool returns a warnings-only response per the
+ * §5.8 graceful-degradation pattern.
+ */
+import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+import { z } from 'zod';
+import { type IndexedRecord } from './storage.js';
+declare const SummarizeInput: z.ZodObject<{
+    context_id: z.ZodOptional<z.ZodString>;
+    record_hashes: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+    focus: z.ZodOptional<z.ZodString>;
+    max_records: z.ZodOptional<z.ZodNumber>;
+    model: z.ZodOptional<z.ZodString>;
+}, "strip", z.ZodTypeAny, {
+    context_id?: string | undefined;
+    model?: string | undefined;
+    record_hashes?: string[] | undefined;
+    focus?: string | undefined;
+    max_records?: number | undefined;
+}, {
+    context_id?: string | undefined;
+    model?: string | undefined;
+    record_hashes?: string[] | undefined;
+    focus?: string | undefined;
+    max_records?: number | undefined;
+}>;
+interface SummarizeOutput {
+    narrative: string | null;
+    cited_record_hashes: string[];
+    records_summarized: number;
+    records_skipped: number;
+    records_with_sidecar: number;
+    records_without_sidecar: number;
+    model_used: string | null;
+    warnings: string[];
+}
+export interface AtribSummarizeServer {
+    mcp: McpServer;
+}
+export declare function createAtribSummarizeServer(): Promise<AtribSummarizeServer>;
+declare function handleSummarize(input: z.infer<typeof SummarizeInput>): Promise<SummarizeOutput>;
+declare function selectRecords(input: z.infer<typeof SummarizeInput>, byHash: Map<string, IndexedRecord>, newestFirst: IndexedRecord[]): IndexedRecord[];
+export declare const __test_only__: {
+    handleSummarize: typeof handleSummarize;
+    selectRecords: typeof selectRecords;
+};
+export {};

package/dist/index.js

@@ -0,0 +1,159 @@
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * atrib-summarize MCP server: registers the `summarize` tool that reads
+ * a set of records (by context_id, by record_hashes, or both) and asks
+ * an OpenAI-compatible LLM to synthesize a narrative across them.
+ *
+ * Closes the consumer-side cognitive-loop primitive companion to trace:
+ * trace returns the causal chain (structural); summarize returns the
+ * synthesized meaning across that chain (semantic). Both read the same
+ * local mirror including the optional _local sidecar.
+ *
+ * LLM access via OpenAI-compatible HTTP. Defaults to NVIDIA NIM with the
+ * Qwen 397B model; override via env vars or per-call model override.
+ * Without an API key, the tool returns a warnings-only response per the
+ * §5.8 graceful-degradation pattern.
+ */
+import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+import { z } from 'zod';
+import { loadAllRecords } from './storage.js';
+import { callLlm, resolveLlmConfig } from './llm.js';
+import { buildSystemPrompt, buildUserMessage } from './prompt.js';
+const SHA256_REF_PATTERN = /^sha256:[0-9a-f]{64}$/;
+const HEX_32_PATTERN = /^[0-9a-f]{32}$/;
+const SummarizeInput = z.object({
+    context_id: z.string().regex(HEX_32_PATTERN).optional().describe("When supplied, summarize every record sharing this 32-hex trace identifier. " +
+        "Mutually composable with record_hashes: if both, the union of records is used."),
+    record_hashes: z.array(z.string().regex(SHA256_REF_PATTERN)).optional().describe("Explicit list of 'sha256:<64-hex>' record hashes to summarize. Records not " +
+        "in the local mirror are silently skipped (use trace to inspect dangling)."),
+    focus: z.string().min(1).max(2000).optional().describe('Optional steering for the synthesis (e.g. "what decisions did I make", ' +
+        '"why did I revise X", "what are the load-bearing claims"). Defaults to a ' +
+        'general "summarize what the agent did and why" focus.'),
+    max_records: z.number().int().min(1).max(200).optional().describe('Cap on records fed to the LLM, default 50. Records beyond the cap (after ' +
+        'sorting by timestamp ascending) are skipped and counted in records_skipped. ' +
+        'Increase carefully — large prompts run slower + cost more + risk model ' +
+        'context limits.'),
+    model: z.string().min(1).max(200).optional().describe('Override the configured model for this call. Format depends on provider ' +
+        '(NVIDIA NIM uses e.g. "qwen/qwen3.5-397b-a17b"). Defaults to env ' +
+        'ATRIB_SUMMARIZE_MODEL or the package default.'),
+});
+export async function createAtribSummarizeServer() {
+    const mcp = new McpServer({ name: 'atrib-summarize', version: '0.1.0' });
+    mcp.registerTool('summarize', {
+        title: 'synthesize narrative across signed records',
+        description: 'Read a set of records (by context_id and/or record_hashes) and ask ' +
+            'an LLM to synthesize a coherent narrative across them. Reads the local ' +
+            'signed-record mirror (~/.atrib/records/*.jsonl) including the optional ' +
+            '_local sidecar for semantic content. Records without a sidecar are ' +
+            'flagged so the LLM can be honest about impoverished input.',
+        inputSchema: SummarizeInput.shape,
+    }, async (rawInput) => {
+        const input = SummarizeInput.parse(rawInput);
+        const result = await handleSummarize(input);
+        return {
+            content: [{ type: 'text', text: JSON.stringify(result, null, 2) }],
+        };
+    });
+    return { mcp };
+}
+async function handleSummarize(input) {
+    const warnings = [];
+    const maxRecords = input.max_records ?? 50;
+    if (!input.context_id && !input.record_hashes) {
+        warnings.push('one of context_id or record_hashes is required');
+        return emptyOutput(warnings);
+    }
+    const llmCfg = resolveLlmConfig(input.model);
+    if (!llmCfg) {
+        warnings.push('no LLM API key resolved (set ATRIB_SUMMARIZE_API_KEY or NVIDIA_API_KEY); cannot synthesize');
+        return emptyOutput(warnings);
+    }
+    // Load local mirror once; filter to the requested set.
+    const { byHash, newestFirst } = loadAllRecords();
+    const selected = selectRecords(input, byHash, newestFirst);
+    if (selected.length === 0) {
+        warnings.push('no records matched the request');
+        return { ...emptyOutput(warnings), model_used: llmCfg.model };
+    }
+    // Sort ascending by timestamp + apply max_records cap.
+    selected.sort((a, b) => a.record.timestamp - b.record.timestamp);
+    const summarized = selected.slice(0, maxRecords);
+    const skipped = selected.length - summarized.length;
+    if (skipped > 0) {
+        warnings.push(`${skipped} record${skipped === 1 ? '' : 's'} skipped beyond max_records=${maxRecords}`);
+    }
+    const withSidecar = summarized.filter((r) => r.local).length;
+    const withoutSidecar = summarized.length - withSidecar;
+    if (withoutSidecar > 0) {
+        warnings.push(`${withoutSidecar} record${withoutSidecar === 1 ? '' : 's'} lacked a _local sidecar; LLM was given only event_type + cryptographic metadata for those`);
+    }
+    const systemMsg = buildSystemPrompt();
+    const userMsg = buildUserMessage(summarized, input.focus);
+    let narrative = '';
+    let modelUsed = llmCfg.model;
+    try {
+        const llmResult = await callLlm(llmCfg, systemMsg, userMsg);
+        narrative = llmResult.content;
+        modelUsed = llmResult.model;
+    }
+    catch (e) {
+        warnings.push(`LLM call failed: ${e instanceof Error ? e.message : String(e)}`);
+        return {
+            narrative: null,
+            cited_record_hashes: summarized.map((r) => r.record_hash),
+            records_summarized: summarized.length,
+            records_skipped: skipped,
+            records_with_sidecar: withSidecar,
+            records_without_sidecar: withoutSidecar,
+            model_used: llmCfg.model,
+            warnings,
+        };
+    }
+    return {
+        narrative,
+        cited_record_hashes: summarized.map((r) => r.record_hash),
+        records_summarized: summarized.length,
+        records_skipped: skipped,
+        records_with_sidecar: withSidecar,
+        records_without_sidecar: withoutSidecar,
+        model_used: modelUsed,
+        warnings,
+    };
+}
+function selectRecords(input, byHash, newestFirst) {
+    const seen = new Set();
+    const out = [];
+    if (input.record_hashes) {
+        for (const h of input.record_hashes) {
+            const idx = byHash.get(h);
+            if (idx && !seen.has(idx.record_hash)) {
+                seen.add(idx.record_hash);
+                out.push(idx);
+            }
+        }
+    }
+    if (input.context_id) {
+        for (const r of newestFirst) {
+            if (r.record.context_id === input.context_id && !seen.has(r.record_hash)) {
+                seen.add(r.record_hash);
+                out.push(r);
+            }
+        }
+    }
+    return out;
+}
+function emptyOutput(warnings) {
+    return {
+        narrative: null,
+        cited_record_hashes: [],
+        records_summarized: 0,
+        records_skipped: 0,
+        records_with_sidecar: 0,
+        records_without_sidecar: 0,
+        model_used: null,
+        warnings,
+    };
+}
+// Test-only export — lets unit tests exercise the assembly logic without
+// going through the McpServer transport.
+export const __test_only__ = { handleSummarize, selectRecords };

package/dist/llm.d.ts

@@ -0,0 +1,29 @@
+/**
+ * OpenAI-compatible chat completion client.
+ *
+ * Defaults to the env-configured endpoint and model. Designed for use
+ * with NVIDIA NIM, Together, OpenRouter, or any OpenAI-compatible
+ * provider. The summarize tool is the only consumer.
+ */
+export interface LlmConfig {
+    baseUrl: string;
+    model: string;
+    apiKey: string;
+    maxTokens: number;
+    temperature: number;
+    timeoutMs: number;
+}
+export interface LlmResult {
+    content: string;
+    model: string;
+}
+/**
+ * Resolve LLM config from env with documented defaults. Caller-supplied
+ * model override (from the MCP tool input) wins over env.
+ */
+export declare function resolveLlmConfig(modelOverride?: string): LlmConfig | null;
+/**
+ * POST one chat completion. Throws on HTTP error so the caller can
+ * surface it via warnings; never silently returns empty content.
+ */
+export declare function callLlm(cfg: LlmConfig, systemMsg: string, userMsg: string): Promise<LlmResult>;

package/dist/llm.js

@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * Resolve LLM config from env with documented defaults. Caller-supplied
+ * model override (from the MCP tool input) wins over env.
+ */
+export function resolveLlmConfig(modelOverride) {
+    const apiKey = process.env['ATRIB_SUMMARIZE_API_KEY'] ??
+        process.env['NVIDIA_API_KEY'] ??
+        process.env['NVIDIA_NIM_API_KEY'] ??
+        '';
+    if (!apiKey)
+        return null;
+    return {
+        baseUrl: process.env['ATRIB_SUMMARIZE_BASE_URL'] ?? 'https://integrate.api.nvidia.com/v1',
+        model: modelOverride ?? process.env['ATRIB_SUMMARIZE_MODEL'] ?? 'qwen/qwen3.5-397b-a17b',
+        apiKey,
+        maxTokens: Number(process.env['ATRIB_SUMMARIZE_MAX_TOKENS'] ?? 4000),
+        temperature: Number(process.env['ATRIB_SUMMARIZE_TEMPERATURE'] ?? 0.3),
+        timeoutMs: Number(process.env['ATRIB_SUMMARIZE_TIMEOUT_MS'] ?? 120000),
+    };
+}
+/**
+ * POST one chat completion. Throws on HTTP error so the caller can
+ * surface it via warnings; never silently returns empty content.
+ */
+export async function callLlm(cfg, systemMsg, userMsg) {
+    const url = cfg.baseUrl.replace(/\/$/, '') + '/chat/completions';
+    const ctrl = new AbortController();
+    const timer = setTimeout(() => ctrl.abort(), cfg.timeoutMs);
+    try {
+        const res = await fetch(url, {
+            method: 'POST',
+            headers: {
+                Authorization: `Bearer ${cfg.apiKey}`,
+                'Content-Type': 'application/json',
+            },
+            body: JSON.stringify({
+                model: cfg.model,
+                messages: [
+                    { role: 'system', content: systemMsg },
+                    { role: 'user', content: userMsg },
+                ],
+                max_tokens: cfg.maxTokens,
+                temperature: cfg.temperature,
+            }),
+            signal: ctrl.signal,
+        });
+        if (!res.ok) {
+            const text = await res.text().catch(() => '<no body>');
+            throw new Error(`LLM POST ${res.status}: ${text.slice(0, 500)}`);
+        }
+        const json = (await res.json());
+        const content = json.choices?.[0]?.message?.content ?? '';
+        if (!content)
+            throw new Error('LLM response had empty content');
+        return { content, model: json.model ?? cfg.model };
+    }
+    finally {
+        clearTimeout(timer);
+    }
+}

package/dist/main.d.ts

@@ -0,0 +1,2 @@
+#!/usr/bin/env node
+export {};

package/dist/main.js

@@ -0,0 +1,16 @@
+#!/usr/bin/env node
+// SPDX-License-Identifier: Apache-2.0
+// atrib-summarize standalone binary. Wires the McpServer to a stdio transport
+// so it can be launched as a subprocess by an MCP host (Claude Code,
+// Claude Desktop, etc.).
+import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
+import { createAtribSummarizeServer } from './index.js';
+async function main() {
+    const { mcp } = await createAtribSummarizeServer();
+    const transport = new StdioServerTransport();
+    await mcp.connect(transport);
+}
+main().catch((e) => {
+    console.error('atrib-summarize: fatal', e instanceof Error ? e.stack ?? e.message : String(e));
+    process.exit(1);
+});

package/dist/prompt.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Prompt construction for the summarize tool.
+ *
+ * Renders a chronological prose digest of N records and feeds it to an
+ * LLM with instructions to synthesize a narrative. The records' semantic
+ * content (from the local sidecar) is included when available; legacy
+ * bare-record entries fall back to event_type + cryptographic metadata
+ * only, with a flag so the LLM knows the input is impoverished.
+ */
+import type { IndexedRecord } from './storage.js';
+export declare function buildSystemPrompt(): string;
+/**
+ * Render the user message: a chronological digest of every record fed
+ * to the LLM, followed by the focus question (if any).
+ */
+export declare function buildUserMessage(records: IndexedRecord[], focus: string | undefined): string;

package/dist/prompt.js

@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: Apache-2.0
+const SYSTEM_PROMPT = `You synthesize narratives across signed agent activity records.
+
+Each record has:
+- A cryptographic identity (signature, hash, creator_key)
+- A causal position (chain_root, informed_by)
+- A semantic payload, EITHER carried in a local sidecar (rich) OR absent (impoverished, only event_type and metadata)
+
+Your job: produce a coherent narrative across the records that surfaces:
+- What the agent was doing (decisions, observations, revisions)
+- What informed what (causal chains via informed_by)
+- What's load-bearing vs incidental
+- Honest gaps (when records are impoverished, say so; do not invent semantics)
+
+Cite specific records by short hash (first 12 chars of record_hash) when making claims.
+Be concise. The reader is the same agent that signed these records — assume they want to be reminded, not lectured.`;
+export function buildSystemPrompt() {
+    return SYSTEM_PROMPT;
+}
+/**
+ * Render the user message: a chronological digest of every record fed
+ * to the LLM, followed by the focus question (if any).
+ */
+export function buildUserMessage(records, focus) {
+    const sortedAsc = [...records].sort((a, b) => a.record.timestamp - b.record.timestamp);
+    const lines = [];
+    lines.push(`# Records to synthesize (${records.length}, oldest-first)`);
+    lines.push('');
+    for (const r of sortedAsc) {
+        lines.push(renderRecord(r));
+        lines.push('');
+    }
+    lines.push('---');
+    lines.push('');
+    if (focus && focus.trim().length > 0) {
+        lines.push(`# Focus`);
+        lines.push(focus.trim());
+    }
+    else {
+        lines.push(`# Default focus`);
+        lines.push('Summarize what the agent did and why, surfacing the load-bearing decisions and the causal chain.');
+    }
+    return lines.join('\n');
+}
+function shortHash(h) {
+    // strip "sha256:" prefix and shorten
+    const noPrefix = h.startsWith('sha256:') ? h.slice(7) : h;
+    return noPrefix.slice(0, 12) + '…';
+}
+function renderRecord(r) {
+    const ts = new Date(r.record.timestamp).toISOString();
+    const evt = r.record.event_type.split('/').pop() ?? r.record.event_type;
+    const hashShort = shortHash(r.record_hash);
+    const ctxShort = r.record.context_id.slice(0, 8) + '…';
+    const informedBy = Array.isArray(r.record.informed_by) ? r.record.informed_by : [];
+    const out = [];
+    out.push(`## ${ts} · ${evt} · ${hashShort} · trace ${ctxShort}`);
+    if (informedBy.length > 0) {
+        out.push(`informed_by: ${informedBy.map(shortHash).join(', ')}`);
+    }
+    const sc = r.local;
+    if (sc) {
+        if (sc.toolName)
+            out.push(`tool: ${sc.toolName}`);
+        const c = sc.content;
+        if (c) {
+            if (typeof c['what'] === 'string')
+                out.push(`what: ${c['what']}`);
+            if (typeof c['why_noted'] === 'string')
+                out.push(`why_noted: ${c['why_noted']}`);
+            if (typeof c['summary'] === 'string')
+                out.push(`summary: ${c['summary']}`);
+            if (typeof c['importance'] === 'string')
+                out.push(`importance: ${c['importance']}`);
+            if (typeof c['prior_position'] === 'string')
+                out.push(`prior_position: ${c['prior_position']}`);
+            if (typeof c['new_position'] === 'string')
+                out.push(`new_position: ${c['new_position']}`);
+            if (typeof c['reason'] === 'string')
+                out.push(`reason: ${c['reason']}`);
+            if (Array.isArray(c['topics'])) {
+                out.push(`topics: ${c['topics'].filter((x) => typeof x === 'string').join(', ')}`);
+            }
+        }
+        // Tool-call sidecar may have args/result; keep brief to control prompt size.
+        if (sc.args)
+            out.push(`args (truncated): ${JSON.stringify(sc.args).slice(0, 300)}`);
+        if (sc.result)
+            out.push(`result (truncated): ${JSON.stringify(sc.result).slice(0, 300)}`);
+    }
+    else {
+        out.push('(no semantic sidecar available — record predates local-mirror sidecar pattern; only cryptographic metadata is present)');
+    }
+    return out.join('\n');
+}

package/dist/storage.d.ts

@@ -0,0 +1,36 @@
+import { type AtribRecord, type OnRecordSidecar } from '@atrib/mcp';
+export interface IndexedRecord {
+    record: AtribRecord;
+    /** sha256:<64-hex> form, matching informed_by entries. */
+    record_hash: string;
+    /** Source file the record was read from (for debugging). */
+    source: string;
+    /**
+     * Pre-sign payload context, when the producer wrote it to the local
+     * mirror as a `_local` sidecar on the envelope. Carries semantic content
+     * (toolName, args, result, content) the signed AtribRecord COMMITS TO
+     * via content_id / args_hash / result_hash but does not itself contain.
+     * Absent on legacy bare-record entries from older mirror writes.
+     */
+    local?: SidecarPayload;
+}
+/**
+ * Combined sidecar shape — superset of every producer's local payload
+ * (mcp-wrap writes toolName/args/result; atrib-emit writes content).
+ */
+export interface SidecarPayload extends OnRecordSidecar {
+    content?: Record<string, unknown>;
+    producer?: string;
+}
+/**
+ * Read every *.jsonl mirror in ATRIB_RECORDS_DIR and return an index keyed
+ * by sha256:<64-hex> record_hash. Both shapes (bare AtribRecord and emit's
+ * { record, proof, written_at } envelope) are supported.
+ *
+ * Returns the indexed map AND a flat list (newest-first by timestamp) for
+ * callers that want recall-style enumeration without by-hash lookup.
+ */
+export declare function loadAllRecords(dir?: string): {
+    byHash: Map<string, IndexedRecord>;
+    newestFirst: IndexedRecord[];
+};

package/dist/storage.js

@@ -0,0 +1,84 @@
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * Local mirror reader for atrib-trace.
+ *
+ * Reads every *.jsonl mirror under ~/.atrib/records/ and normalizes to bare
+ * AtribRecord shape regardless of producer (wrapper-signed or emit envelope).
+ *
+ * Builds an in-memory index by record_hash so trace can walk informed_by
+ * chains in O(1) per hop. The hash key is sha256:<64-hex> (the same form
+ * informed_by entries use), so no transformation needed at lookup time.
+ */
+import { existsSync, readFileSync, readdirSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { join } from 'node:path';
+import { canonicalRecord, hexEncode, sha256 } from '@atrib/mcp';
+const RECORDS_DIR = process.env.ATRIB_RECORDS_DIR ?? join(homedir(), '.atrib', 'records');
+/**
+ * Read every *.jsonl mirror in ATRIB_RECORDS_DIR and return an index keyed
+ * by sha256:<64-hex> record_hash. Both shapes (bare AtribRecord and emit's
+ * { record, proof, written_at } envelope) are supported.
+ *
+ * Returns the indexed map AND a flat list (newest-first by timestamp) for
+ * callers that want recall-style enumeration without by-hash lookup.
+ */
+export function loadAllRecords(dir = RECORDS_DIR) {
+    const byHash = new Map();
+    const all = [];
+    if (!existsSync(dir))
+        return { byHash, newestFirst: [] };
+    let files = [];
+    try {
+        files = readdirSync(dir).filter((f) => f.endsWith('.jsonl'));
+    }
+    catch {
+        return { byHash, newestFirst: [] };
+    }
+    for (const fname of files) {
+        const path = join(dir, fname);
+        try {
+            const raw = readFileSync(path, 'utf8');
+            for (const line of raw.split('\n')) {
+                const trimmed = line.trim();
+                if (!trimmed)
+                    continue;
+                try {
+                    const parsed = JSON.parse(trimmed);
+                    // Normalize: envelope shape vs legacy bare record shape.
+                    const isEnvelope = 'record' in parsed &&
+                        parsed.record &&
+                        parsed.record.context_id !== undefined;
+                    const rec = isEnvelope
+                        ? parsed.record
+                        : parsed;
+                    if (!rec.context_id || !rec.signature || !rec.timestamp)
+                        continue;
+                    const hashHex = hexEncode(sha256(canonicalRecord(rec)));
+                    const indexed = {
+                        record: rec,
+                        record_hash: `sha256:${hashHex}`,
+                        source: fname,
+                    };
+                    // Lift the `_local` sidecar onto the indexed record when present.
+                    // Legacy bare-record entries have no sidecar; that's OK — consumers
+                    // tolerate its absence.
+                    if (isEnvelope) {
+                        const sidecar = parsed._local;
+                        if (sidecar)
+                            indexed.local = sidecar;
+                    }
+                    byHash.set(indexed.record_hash, indexed);
+                    all.push(indexed);
+                }
+                catch {
+                    // Malformed line; skip.
+                }
+            }
+        }
+        catch {
+            // File read failure; skip this file.
+        }
+    }
+    all.sort((a, b) => b.record.timestamp - a.record.timestamp);
+    return { byHash, newestFirst: all };
+}

package/package.json

@@ -0,0 +1,34 @@
+{
+  "name": "@atrib/summarize",
+  "version": "0.2.0",
+  "description": "MCP server for atrib. Synthesizes a narrative across N records via an OpenAI-compatible LLM so agents read context, not raw record bytes.",
+  "license": "Apache-2.0",
+  "type": "module",
+  "main": "./dist/index.js",
+  "bin": {
+    "atrib-summarize": "./dist/main.js"
+  },
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.29.0",
+    "zod": "^3.25.76",
+    "@atrib/mcp": "0.4.0"
+  },
+  "devDependencies": {
+    "@types/node": "^22.19.17",
+    "tsx": "^4.21.0",
+    "typescript": "^5.9.3",
+    "vitest": "^3.2.4"
+  },
+  "files": [
+    "dist"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "scripts": {
+    "build": "rm -rf dist && tsc",
+    "start": "node dist/main.js",
+    "dev": "tsx src/main.ts",
+    "test": "vitest run"
+  }
+}