llms-txt

Claude Code Chrome Extension

Platform-specific operational guidance for using Claude Code through the Chrome extension, with attention to message composition, app invocation, list-formatting hazards, and ceremonial prompt decomposition.


Purpose

This document captures evaluated elevation: guidance derived from lived use, observed failure modes, and corrections that should become reusable protocol rather than one-off troubleshooting.

The aim is not merely to “use the tool correctly,” but to preserve conversational continuity, prevent avoidable fragmentation, and ensure that ceremonial preparation happens before complex execution begins.


Core Principle

Treat platform mechanics as part of the work.

A prompt can fail before the model reasons about it. UI behavior, message submission rules, mention mechanics, and editor formatting all shape whether the intended instruction arrives intact.

Therefore:

  1. Orient before acting
  2. Respect the platform’s actual interaction model
  3. Convert repeated mistakes into explicit guidance
  4. Elevate lived corrections into portable protocol

Ceremonial Opening: PDE First

Before executing any complex multi-step request, always decompose it first using the Prompt Decomposer:

http://etuaptmumk.vercel.app/prompt-decomposer

This is the ceremonial opening.

Why this comes first

Prompt decomposition:

Within this guidance set, PDE is not optional preamble. It is an orientation ritual that turns a request into an actionable structure.

Operational rule

Use PDE first when the request involves any of the following:

For trivial one-step actions, direct execution may be fine. For everything else: decompose before doing.

See also: Prompt Decomposition Engine (PDE)


Platform Corrections Log

The entries below follow the evaluated elevation pattern:


1. Slack message composition fragments on Enter

Context

Slack message composition while drafting long, multi-line instructions to an app or agent.

Failure mode

Pressing Enter sends the message immediately instead of inserting a newline. A long intended message can become fragmented across multiple Slack messages. This breaks the conversational thread and may prevent the receiving app from being notified about the full content. In practice, some apps only receive the first fragment as the actionable notification.

Correction

In Slack, use Shift+Enter to create newlines while composing a single multi-line message. Before sending a long instruction, ensure the full prompt remains in one message body.

Why it matters

A fragmented prompt is not merely untidy; it changes system behavior. The receiving app may treat only the first fragment as canonical, causing lost requirements, broken intent transmission, and response drift.

Elevation status

Elevated from lived session failure into standing protocol.

Applies to


2. Slack app mentions must use autocomplete

Context

Mentioning an app inside Slack so it is properly addressed and notified.

Failure mode

Typing an app name manually without selecting it from Slack’s autocomplete dropdown may produce text that looks like a mention but does not function as one.

Correction

When mentioning an app in Slack, use the autocomplete dropdown and select the app explicitly rather than typing the name freehand.

Why it matters

Correct invocation depends on Slack recognizing the mention as an actual platform object, not as plain text. If the mention is malformed, the intended app may not be triggered.

Elevation status

Elevated from platform-specific invocation failure.

Applies to


3. Slack apps are not added like people

Context

Adding an app such as MiaOpenClaw to a Slack channel.

Failure mode

Attempting to add an app through Slack’s “Add people to channel” dialog, as though the app were a human participant.

Correction

Treat Slack apps as integrations, not people. Add them via:

Do not rely on the people-invite flow for app onboarding.

Why it matters

This is a category error at the platform layer. If the app is treated as a person, setup fails before collaboration even begins.

Elevation status

Elevated from failed onboarding attempt.

Applies to


4. GitHub and Simplenote list continuation can distort plain text

Context

Writing notes, issue content, or guidance in editors used by GitHub and Simplenote.

Failure mode

When a line begins with a dash or asterisk, pressing Enter may continue list formatting automatically. This can unintentionally transform plain text into list structure or extend a list when the intention was to return to normal prose.

Correction

Be deliberate after list items in GitHub and Simplenote. Watch for auto-continuation behavior after - or *. If plain text is intended, explicitly break out of the list rather than assuming a single Enter will reset formatting.

Why it matters

Formatting drift changes meaning, readability, and sometimes downstream parsing. In instruction-heavy work, accidental list continuation can deform the conceptual structure of the guidance.

Elevation status

Elevated from repeated editor-behavior correction.

Applies to


Collaboration Insight: Ceremony over A2A

A notable meta-observation from this work:

Two instances may collaborate effectively without a formal agent-to-agent protocol when they share ceremony, orientation, and explicit guidance structures.

In this frame, collaboration does not begin with transport. It begins with:

This suggests that ceremonial coherence can partially substitute for protocol-level integration.

That insight belongs not only to platform guidance but also to broader reflection on companion architectures, memory-aware collaboration, and narrative framing.


When using Claude Code through the Chrome extension in a cross-platform workflow:

  1. Run PDE first for any complex ask
  2. Draft the full prompt as one coherent message
  3. In Slack, use Shift+Enter for line breaks
  4. Use real app mentions via autocomplete
  5. Confirm the app is actually present in the channel via /invite @AppName or Integrations if needed
  6. Watch editor formatting behavior when copying guidance into GitHub or Simplenote
  7. Promote recurring corrections into durable guidance rather than solving them repeatedly from scratch

Why This Document Exists

This is not generic documentation.

It is a vessel for mistakes that became method.

Every entry here comes from the recognition that a tiny platform behavior can derail a sophisticated collaboration. The corrective act is to turn friction into doctrine, doctrine into reusable guidance, and reusable guidance into better ceremony.

That is evaluated elevation.


See Also