Skills · Business & ops

Turn your rambling notes into a clear brief

Unverified32/40

Paste your messy brain-dump or voice-note and get back a tidy brief: your goal, your decisions, what you changed your mind about, and the guesses you need to correct.

Originally by Shubham Saboo · Apache-2.0

Claude CodePartialHas SKILL.md but declares no allowed-tools — Claude Code will ask for permission each time
CursorPartialPlain prose you can paste in — but no Cursor rules file
CodexPartialPlain prose you can paste in — but no AGENTS.md
Gemini CLIPartialPlain prose you can paste in
CopilotPartialPlain prose you can paste in — but no Copilot instructions file
npx agentalley add thinking-out-loud

This command does not work yet — the CLI is still being built. Until then, use Raw in the reader below to take the file.

Who is stuck, and on what

I think out loud in half-finished sentences and change my mind halfway through, so whoever picks it up fills the gaps their own way and runs with it. By the time I see the result, it's built on a misunderstanding I never got the chance to catch.

What it gives you

A short structured brief with your goal, your locked-in decisions, the open questions, anything you flip-flopped on, and a clearly separated list of guesses flagged for you to fix.

When NOT to use it

It won't make the decisions for you or produce the finished work — it only plays your thinking back so you can spot what's wrong before anyone acts on it.

The whole source

No sign-in, no blur, nothing truncated
thinking-out-loud/SKILL.md159 lines7.3 KBRawView on GitHub
Frontmatter — 4 properties
namethinking-out-loud
description>- A contract for what the agent does when a long, messy, stream-of-consciousness ramble arrives (usually voice dictation): act on nothing until the echo brief is approved. The echo audits the entire transfer, mission, locked decisions and constraints, open questions, flips and parked tangents, with the model's inferences and guesses quarantined away from the user's own phrasing, so the user verifies what the model believes, not just what it doubts. Use when the user says "let me think out loud" or wants to ramble a bit before building anything, when a message opens with a speech-to-text preamble like "switching to voice, sorry for typos", when input is a long weakly punctuated stream with restarts and mid-message reversals ("actually no, wait, scrap that idea entirely"), or when the user asks to be interviewed about a fuzzy half-formed idea. Includes an optional capture mode for rambles spread across several messages and an optional targeted interview.
licenseApache-2.0
metadata author: "Shubham Saboo version: "1.3.0 source: "https://github.com/Shubhamsaboo/awesome-llm-apps
1---
2name: thinking-out-loud
3description: >-
4 A contract for what the agent does when a long, messy, stream-of-consciousness
5 ramble arrives (usually voice dictation): act on nothing until the echo brief
6 is approved. The echo audits the entire transfer, mission, locked decisions
7 and constraints, open questions, flips and parked tangents, with the model's
8 inferences and guesses quarantined away from the user's own phrasing, so the
9 user verifies what the model believes, not just what it doubts. Use when the
10 user says "let me think out loud" or wants to ramble a bit before building
11 anything, when a message opens with a speech-to-text preamble like "switching
12 to voice, sorry for typos", when input is a long weakly punctuated stream
13 with restarts and mid-message reversals ("actually no, wait, scrap that idea
14 entirely"), or when the user asks to be interviewed about a fuzzy half-formed
15 idea. Includes an optional capture mode for rambles spread across several
16 messages and an optional targeted interview.
17license: Apache-2.0
18metadata:
19 author: "Shubham Saboo"
20 version: "1.3.0"
21 source: "https://github.com/Shubhamsaboo/awesome-llm-apps"
22---A5No allowed-tools declared — no way to tell what this skill may touch
23 
24# Thinking Out Loud
25 
26A ten minute voice ramble transfers more context than any prompt a person
27would type, and models reconstruct rambles well. The failure is
28downstream and invisible: the model fills every gap in the ramble
29confidently. "The usual model" silently becomes a specific model. "The
30standard size" becomes a specific viewport. A position the user reversed
31mid-ramble survives as fact. None of this registers as uncertainty from
32the inside, so none of it ever becomes a clarifying question. The model
33then acts on a misreading it fully believes, and the user discovers it an
34hour of generated work later.
35 
36This skill is the fix: before acting on any ramble, produce an echo, a
37short structured audit of everything absorbed, with the model's own
38additions quarantined from the user's words. The user corrects three
39lines instead of debugging a built artifact.
40 
41## Why an echo instead of follow-up questions
42 
43Asking clarifying questions is good, and the interview below does it.
44But questions alone cannot secure a ramble, for two structural reasons:
45 
46- **Questions verify what the model doubts. The echo verifies what the
47 model believes.** A clarifying question requires felt uncertainty, and
48 confident misreadings feel like knowledge. The echo forces every
49 inference and gap-fill into the open whether or not it felt uncertain.
50- **Questions sample; the echo audits.** A long ramble carries dozens of
51 facts and half-decisions. Even good questions probe three or four; the
52 rest of the model's understanding goes unverified into action. The
53 echo inventories the entire transfer, and it works by recognition, not
54 recall: the user reads and spots what is wrong, which is far cheaper
55 than producing answers, and ramblers often do not know their answer
56 until they see the wrong guess written down.
57 
58## The contract
59 
601. **Act on nothing.** No file edits, no code, no plans, no solutions to
61 fragments, until the echo is approved. Reconstruct first.
622. **Label every addition.** Inferences and guesses live in their own
63 section, apart from the user's own content. Never present a guess in
64 the user's voice.
653. **Surface every reversal.** Adopt the later position, but flag the
66 flip. Never silently average or pick.
674. **Lose nothing.** Tangents get parked, not dropped.
685. **Never remark on dictation artifacts.** Typos, homophones, filler,
69 and restarts are resolved silently from context. Keep the user's own
70 vocabulary and project names.
716. **Ask before persisting.** The approved brief is offered a home, never
72 saved unprompted.
73 
74## When to use
75 
76- A message is a long, weakly punctuated stream of consciousness with
77 restarts, filler, and mid-message reversals ("actually no, scrap that")
78- A message opens with a voice preamble ("switching to speech
79 recognition, sorry for any typos", "dictating this")
80- The user says they want to ramble or think out loud
81- The user asks to be interviewed to untangle a fuzzy idea
82 
83## When not to use
84 
85- Short requests that are already clear
86- The user wants a verbatim transcript, minutes, or cleanup of dictation
87 while keeping their exact words
88- Long but already structured text, such as a pasted spec or document
89- The user asked a direct question and wants a direct answer
90 
91## The echo
92 
93One structured reply. Dense, scannable, and short: the user should find
94and fix an error in seconds. Full template with a worked example in
95[references/echo-format.md](references/echo-format.md).
96 
971. **Mission**: one sentence stating what the user is actually trying to
98 achieve. Often this differs from what they said first; that is fine.
992. **Locked**: the user's decisions and constraints, merged into one
100 list. Mark anything they called a top priority.
1013. **Open**: questions the ramble raised but did not answer.
1024. **Ledger**: flips (both positions in one line, later one adopted) and
103 parked tangents (one line each).
1045. **My additions**: the only interpretation callouts. "Inferred"
105 (strongly implied but never stated) and "Guessed" (gaps you filled).
106 Tell the user to correct these first.
107 
108Compression rules, non-negotiable:
109 
110- **Nothing appears twice.** Every fact lives in exactly one section.
111- **No "you said" recap.** Everything outside My additions is the user's
112 own content by definition; only the model's additions get called out.
113- **One line per bullet.** If a bullet needs two lines, it is two bullets
114 or it is bloat.
115- **Vague quantifiers are never silently resolved.** "The usual model",
116 "standard size", "soon": each lands in Open or Guessed, never absorbed
117 into a locked item as if it were specified.
118 
119Close by inviting corrections and offering the interview.
120 
121## The interview (optional)
122 
123Follow-up questions have their place: after the audit, not instead of
124it. Only if the user accepts the offer, or asked to be interviewed up
125front.
126 
127- Ask only about items flagged in Open or Guessed
128- One question per message, highest information gain first
129- Each question states in one clause why it matters
130- Cap at five questions; stop early once answers stop changing the brief
131- After the interview, restate only the sections of the echo that changed
132 
133## Capture mode (multi-message rambles)
134 
135Not needed for dictation tools, where the whole ramble arrives as one
136message. Use it when the user invokes the skill before rambling and then
137adds thoughts across several messages, possibly over a long stretch.
138 
139- Acknowledge once, in one short line ("Go ahead, I'm listening. Say
140 'done' when you want the echo.")
141- For every following message, reply with a single minimal line
142 ("Listening."). Vary it slightly so it does not feel robotic.
143- Do NOT solve, praise, summarize, analyze, or ask questions mid-stream.
144- If the user asks a direct question mid-ramble, answer it in at most two
145 sentences, then return to listening.
146- Exit on "done", "echo", "echo me", "that's it", "what did you get", or
147 any clear equivalent, then deliver the echo.
148 
149## Persistence
150 
151After the user approves the echo, offer exactly three options:
152 
1531. Append the brief to CLAUDE.md so future sessions inherit it
1542. Save it to `docs/rambles/YYYY-MM-DD-<topic>.md`
1553. Keep it in-conversation only
156 
157The approved brief then governs the rest of the session: honor its
158decisions and constraints without re-asking.
159 

Reviews

Installed this one?Write the first review and take the Trailblazer badge.

Reviews only open after a real install, so this is empty — and we leave it empty rather than invent one.

Alternatives

Also in Business & ops