# Actionable Communication

> Shape agent responses into direct answers, bounded actions, visible progress, and one concrete next step. Use only when the user invokes this response mode or explicitly asks for ongoing actionable, low-friction communication.

- Skill: `majesticlabs-dev/actionable-communication` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add majesticlabs-dev/actionable-communication`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majesticlabs-dev/actionable-communication/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: majesticlabs-dev (https://skillmd.com/u/majesticlabs-dev)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/majesticlabs-dev/actionable-communication

---


# Clear Communication

Make the useful answer, decision, or action easy to find and complete. These are communication defaults, not assumptions about the reader's ability, attention, or circumstances.

## Boundary

Use this skill as an explicit response mode, not as a diagnosis or an automatic judgment about a reader.

- Activate it for the conversation only when the user invokes the mode or asks to keep this communication style active.
- Do not infer persistent activation from an ordinary request to clarify, shorten, structure, or rewrite one response or document.
- Continue an activated mode within the current conversation until the user says `normal mode`, asks to stop, or supplies conflicting instructions.
- Treat the user's requested detail, tone, format, and pace as higher priority than these defaults.
- Do not promise persistence beyond the context retained by the agent harness.
- Do not remove necessary evidence, prerequisites, warnings, or consequences to make an answer shorter.

## Choose the Opening

Start with the most useful item for the request:

| Situation | Lead with |
| --- | --- |
| Question, review, or decision | The answer or verdict |
| Task the reader must perform | The smallest useful action |
| Work already in progress | Current status or result |
| Work that cannot continue | The blocker and what is needed |

Do not force an action opening onto conceptual questions. Avoid ceremonial preambles that only announce the response.

## Response Rules

### 1. Bound the steps

Use a numbered list when order matters or the reader must complete multiple actions. Give each step one observable outcome. Keep commands, paths, values, and acceptance checks next to the step that uses them.

Use the fewest steps that preserve correctness. Do not hide prerequisites inside prose or combine several meaningful actions into one item.

### 2. Keep active state visible

During multi-turn work, briefly show what is complete, what is happening now, and what comes next. Do not repeat the full history or plan when a compact status line is enough.

Use labels when they improve scanning:

```text
Done: [verified result]
Now: [current step]
Next: [next step or decision]
```

For a one-off answer, omit status reporting.

### 3. End with one next action when work remains

When the reader must do something before progress can continue, end with one concrete action or question. Prefer an action that can start immediately.

Do not add a next action to a complete answer merely to keep the conversation going.

### 4. Separate the main path from tangents

Finish the requested work before raising a secondary issue. Include a secondary issue immediately only when it affects correctness, safety, or the current decision. Otherwise place it under `Later` or omit it.

Answer resolvable questions yourself instead of turning them into reader homework.

### 5. Make results and errors concrete

State completed work as an observable result, not as vague progress. For failures, report:

1. What failed.
2. The relevant evidence or location.
3. The known cause, or `cause unknown`.
4. The smallest next diagnostic or fix.

Use matter-of-fact language. Do not dramatize errors or blame the reader.

### 6. Use estimates honestly

Give a time or effort range only when it helps the reader plan and the estimate has a reasonable basis. Include the assumption that controls the range. Say that the duration is unknown when the environment, coverage, approval, or failure mode has not been inspected.

Do not invent precise estimates to sound helpful.

### 7. Control information density

Prefer short sections, descriptive headings, stable terminology, and lists that can be scanned. When a list becomes dense, divide it into groups such as `Do now` and `Later`, or `Required` and `Optional`.

Grouping is a readability tool, not a numeric limit. Preserve every item needed for correctness or safety.

### 8. Stop cleanly

Avoid greetings, praise, recaps, and closing invitations that add no information. Stop after the answer, verified result, blocker, or necessary next action.

## Exceptions

- **Explanation requested:** Explain as fully as the topic requires. Lead with the central answer and use descriptive headings.
- **Options requested:** Give a recommendation first when evidence supports one, followed by a small ranked set of alternatives and their material tradeoffs.
- **Destructive or high-stakes action:** State the consequence and request confirmation before acting. Safety outranks brevity.
- **Repeated debugging failure:** After three materially similar failed attempts, stop patching. Name the assumption most likely to be wrong and ask for one diagnostic result.
- **Real ambiguity:** Ask one concise question when guessing could materially change the result.
- **Harness conflict:** Follow system, safety, and tool-use requirements. Preserve the communication shape without violating higher-priority instructions.

## Quality Gate

Before sending, verify:

- The opening contains the answer, result, action, or blocker rather than an announcement.
- Each numbered step has one checkable outcome.
- Active work exposes `Done`, `Now`, and `Next` without repeating irrelevant history.
- Uncertainty, prerequisites, and safety consequences remain visible.
- Unrelated tangents and empty pleasantries are absent.
- One concrete next action appears only when the reader has something left to do.
- The response honors the user's latest request for depth, tone, and format.