Skip to content

The cross-harness, self-improving memory layer for AI agents. Join our community: https://discord.com/invite/zdNChS2fBu

Document Claude Cowork normalization in v1.2.7

shukieshah8c9ea92b1cd1
diff --git a/README.md b/README.md
index 76a9715a..4c9812dd 100644
--- a/README.md
+++ b/README.md
@@ -88,7 +88,7 @@ to compare it with the managed path.
| --- | --- | --- |
| [Antigravity CLI](https://docs.asymptotelabs.ai/cli/supported-runtimes-antigravity-cli) | Native hooks | Prompt, pre-tool, post-tool, stop, invocation, command, and file |
| [Claude Code](https://docs.asymptotelabs.ai/cli/supported-runtimes-claude-code) | Local OTLP export plus optional hooks | Prompt, command, tool, file, approval, API/model lifecycle, MCP connection, subagent, and session |
-| [Claude Cowork](https://docs.asymptotelabs.ai/cli/supported-runtimes-claude-cowork) | Admin-configured OTLP | Prompt, command, tool, and file |
+| [Claude Cowork](https://docs.asymptotelabs.ai/cli/supported-runtimes-claude-cowork) | Admin-configured OTLP | Prompt, assistant response, approval, command, tool, file, MCP, session, model, token usage, and runtime-reported cost |
| [Cline](https://docs.asymptotelabs.ai/cli/supported-runtimes-cline) | Managed plugin hooks | Prompts, task lifecycle and errors, tool lifecycle and results, commands with exit codes, file reads/edits with diffs, MCP activity, and token usage/cost |
| [Codex CLI](https://docs.asymptotelabs.ai/cli/supported-runtimes-codex-cli) | Local OTLP plus a session identity hook | Session, prompt, approval, tool results, and per-user/session/model turn token usage |
| [Cursor](https://docs.asymptotelabs.ai/cli/supported-runtimes-cursor) | Native hooks | Prompt, tool, shell command, MCP-like activity, approval, and file edits |
diff --git a/cli/beacon/README.md b/cli/beacon/README.md
index 32805f6e..3762bdbb 100644
--- a/cli/beacon/README.md
+++ b/cli/beacon/README.md
@@ -585,6 +585,12 @@ Claude Cowork monitoring is configured in the Claude admin console at
by Claude Cowork, so use a durable public HTTPS Collector endpoint for ongoing
monitoring. The `--ngrok` mode is for short-lived local testing and prints an
authenticated tunnel URL plus the matching `Authorization` header.
+Cowork uses the shared Claude normalizer for prompts, assistant responses,
+approvals, commands, files, MCP activity, session lifecycle, model usage,
+tokens, and runtime-reported cost. A Desktop-started task may still carry
+`origin=cloud` when `cowork.surface=remote`; endpoint hostname and OS identify
+the Beacon collector machine. Cowork may expose only an opaque `user.id`, which
+Beacon preserves without inventing a display identity.
OpenClaw Gateway monitoring is configured in OpenClaw's local Gateway config
with the `diagnostics-otel` plugin enabled. Beacon prints a local OTLP/HTTP
diff --git a/docs/changelog/cli.mdx b/docs/changelog/cli.mdx
index dd8bfe51..4bc1ea6c 100644
--- a/docs/changelog/cli.mdx
+++ b/docs/changelog/cli.mdx
@@ -11,6 +11,29 @@ Stay up to date with the latest changes to the Agent Beacon CLI.
## August 2026
+### v1.2.7 · August 31, 2026
+
+- **Claude Cowork uses the shared Claude normalizer** · Cowork OTLP logs now
+ produce the same canonical prompt, assistant response, approval, command,
+ file, MCP, session, model, token, and runtime-reported cost fields as Claude
+ Code when Anthropic emits those attributes. Tool parameters are decoded into
+ structured command and MCP fields instead of remaining only in raw OTLP
+ attributes.
+- **Cowork provenance stays coherent** · Remote Cowork execution is marked
+ `origin=cloud` even when the task was started in Claude Desktop. The source's
+ Cowork user identifier replaces the local collector process identity without
+ mixing a cloud account with a local UID.
+- **Richer retained content** · User prompts and assistant responses are
+ promoted into the canonical GenAI message fields with content hashes and byte
+ counts, while the original OTLP attributes remain available under
+ `raw.attributes`.
+- **Local dashboard refresh** · The loopback dashboard has a clearer navigation
+ and investigation layout, richer runtime identity, session-state filtering,
+ and improved event detail views.
+- **Architecture and positioning updates** · The product overview and
+ architecture guides now describe Beacon around its capture, normalization,
+ storage, detection, and forwarding pipeline stages.
+
### v1.2.5 · August 28, 2026
- **Turn-level usage** · Beacon now configures Codex's local trace exporter and
diff --git a/docs/cli/claude-cowork.mdx b/docs/cli/claude-cowork.mdx
index fb5ed831..56b7d46c 100644
--- a/docs/cli/claude-cowork.mdx
+++ b/docs/cli/claude-cowork.mdx
@@ -62,6 +62,23 @@ Recommended production shape:
- Forward OTLP HTTP paths such as `/v1/logs`, `/v1/metrics`, and `/v1/traces` to the Collector's local `4318` receiver
- Treat Cowork telemetry as sensitive because prompt text, tool parameters, file paths, user email addresses, model usage, and errors may be present before Beacon redaction or export
+## How Cowork events are represented
+
+Beacon applies the same Claude event normalizer to Claude Cowork and Claude
+Code. When Anthropic emits the relevant attributes, Cowork prompts, assistant
+responses, approvals, commands, files, MCP activity, model usage, token counts,
+and runtime-reported cost use the canonical Beacon fields while the source OTLP
+attributes remain under `raw.attributes`.
+
+The environment describes where the task executes, not which application opened
+it. A task started in Claude Desktop can report `cowork.surface=remote`; Beacon
+records that as `origin=cloud`. Endpoint hostname and operating system fields
+identify the machine running the Beacon collector.
+
+Cowork may provide only an opaque `user.id` rather than an email address. Beacon
+preserves the identifier but does not attempt to reverse it or synthesize a
+display name.
+
## Local testing with ngrok
For demos, validation, or local development only, Beacon can create a temporary authenticated ngrok tunnel to the local OTLP HTTP receiver:
diff --git a/docs/concepts/faq.mdx b/docs/concepts/faq.mdx
index fba32f6d..d7725a0b 100644
--- a/docs/concepts/faq.mdx
+++ b/docs/concepts/faq.mdx
@@ -189,10 +189,16 @@ surfaces the provider exposes.
Claude Cowork is configured in the Claude organization admin settings. Anthropic's
service exports OTLP telemetry to a durable collector endpoint, and Asymptote
-normalizes the received events into the shared event model.
+normalizes the received events through the same Claude event pipeline used for
+Claude Code. Supported source attributes become canonical prompts, assistant
+responses, approvals, commands, files, MCP activity, model usage, token counts,
+and runtime-reported cost.
Claude Cowork is not a local hook integration. Production deployments should use
a durable HTTPS collector endpoint rather than a laptop or temporary tunnel.
+A task can still be marked as cloud execution when it was started from Claude
+Desktop because Cowork runs that task remotely. The event's endpoint fields
+identify the Beacon collector machine.
### Where should a cloud AI tenant send OTLP telemetry?
diff --git a/docs/runtimes/claude-cowork.mdx b/docs/runtimes/claude-cowork.mdx
index 599aad85..f74c92a4 100644
--- a/docs/runtimes/claude-cowork.mdx
+++ b/docs/runtimes/claude-cowork.mdx
@@ -19,7 +19,16 @@ Before enabling Claude Cowork telemetry, make sure:
## Collection path
-Claude Cowork exports OTLP telemetry from Anthropic's service to a customer-managed collector endpoint. After events reach Asymptote's collector path, Asymptote writes normalized endpoint events to the runtime JSONL log and makes them available in the local dashboard.
+Claude Cowork exports OTLP telemetry from Anthropic's service to a
+customer-managed collector endpoint. After events reach Beacon's collector,
+Beacon applies the shared Claude normalizer, writes canonical endpoint events to
+the runtime JSONL log, and makes them available in the local dashboard.
+
+Cowork can be opened from Claude Desktop while the task itself runs on
+Anthropic's remote infrastructure. When Anthropic reports
+`cowork.surface=remote`, Beacon records `origin=cloud`. The event's endpoint
+hostname and operating system still identify the machine running the Beacon
+collector, not the remote Cowork execution host.
## Discovery and status
@@ -39,8 +48,13 @@ Claude Cowork is not hook-based.
| Area | Support |
|------|---------|
-| Prompt telemetry | Supported when emitted through Claude Cowork OTLP |
-| Command, tool, and file telemetry | Supported when emitted through Claude Cowork OTLP |
+| Prompt and assistant response telemetry | Retained in canonical GenAI input and output message fields when emitted through Cowork OTLP |
+| Command, tool, and file telemetry | Tool parameters and results normalize into canonical command, tool, and file fields |
+| Approval telemetry | Cowork tool decisions normalize into observed approval requests, allows, and denials |
+| MCP telemetry | MCP connections and tool calls include canonical server and tool names |
+| Model and usage telemetry | Model, token usage, and runtime-reported cost normalize into `gen_ai.usage` |
+| Session and runtime telemetry | Session, API, hook lifecycle, refusal, and error records remain available for investigation |
+| User and origin context | Cowork's source-provided user identifier is preserved; remote Cowork surfaces are marked as cloud execution |
| Local JSONL and dashboard | Supported after events reach the collector |
| MDM deployment | Uses a durable customer-managed HTTPS collector endpoint rather than local MDM hook config |
diff --git a/docs/runtimes/index.mdx b/docs/runtimes/index.mdx
index 340d8cc6..bd72f634 100644
--- a/docs/runtimes/index.mdx
+++ b/docs/runtimes/index.mdx
@@ -45,7 +45,7 @@ Two rows in that table are not hook or OTLP integrations. [fx](/runtimes/vercel-
| Agent harness | Collection path | Telemetry coverage |
|----------------|-----------------|--------------------|
-| [Claude Cowork](/runtimes/claude-cowork) | Admin-configured OTLP | Prompt, command, tool, and file telemetry when emitted through Claude Cowork OTLP |
+| [Claude Cowork](/runtimes/claude-cowork) | Admin-configured OTLP | Prompt, assistant response, approval, command, tool, file, MCP, session, model, token usage, and runtime-reported cost telemetry when emitted through Cowork OTLP |
| [Hermes Agent](/runtimes/hermes-agent) | Shell hooks | Prompt, observed tool, command, file, approval request and response, session lifecycle, and subagent stop telemetry |
| [OpenClaw Gateway](/runtimes/openclaw-gateway) | Gateway-configured OTLP/HTTP | OTLP logs, traces, and metrics from the Gateway diagnostics plugin |
diff --git a/docs/security/data-inventory.mdx b/docs/security/data-inventory.mdx
index b2c65023..c1381c82 100644
--- a/docs/security/data-inventory.mdx
+++ b/docs/security/data-inventory.mdx
@@ -22,7 +22,7 @@ Beacon writes normalized endpoint events only when supported runtimes provide te
| Devin Desktop | Cascade/Windsurf hooks | Prompt, command, MCP tool, file read, and file write telemetry where Desktop exposes Cascade hook payloads |
| Factory Droid | OTLP HTTP plus optional hooks | Session, prompt, write/edit/create tool use, stop, session-end, and available OTLP telemetry |
| Cursor | Native hooks | Prompt, tool, shell command, MCP-like, approval, and file edit telemetry |
-| Claude Cowork | Admin-configured OTLP | Prompt, command, tool, and file telemetry when emitted through Claude Cowork OTLP |
+| Claude Cowork | Admin-configured OTLP | Prompt and assistant response content; approvals; command, tool, file, and MCP activity; session lifecycle; model and token usage; runtime-reported cost; source user identifier and cloud origin when emitted through Cowork OTLP |
| OpenClaw Gateway | Gateway-configured OTLP/HTTP | OTLP logs, traces, and metrics from the Gateway diagnostics plugin |
| [Claude.ai](/runtimes/claude-web) (`claude_web`) | Managed browser extension over local OTLP | Prompt text, assistant response text, request and response model, tool call id/name/arguments/results, token usage, and conversation id from the claude.ai chat stream |
| [ChatGPT](/runtimes/chatgpt-web) (`chatgpt_web`) | Managed browser extension over local OTLP | Prompt text, assistant response text, model, tool call activity, and conversation id from the chatgpt.com chat stream. ChatGPT web does not report token usage in its stream |