diff --git a/Anthropic/claude.ai-human-readable.md b/Anthropic/claude.ai-human-readable.md index 34f6839..ee2c421 100644 --- a/Anthropic/claude.ai-human-readable.md +++ b/Anthropic/claude.ai-human-readable.md @@ -4,6 +4,8 @@ > **Date in prompt:** Monday, March 30, 2026 > **Model:** Claude Opus 4.6 (Claude 4.6 model family) +> **Note:** Sections are ordered to match the actual system prompt sequence. + --- ## Table of Contents @@ -27,6 +29,7 @@ - [17. Network Configuration](#17-network-configuration) - [18. Filesystem Configuration](#18-filesystem-configuration) - [19. Reasoning & Thinking (post-prompt)](#19-reasoning--thinking-post-prompt) +- [20. Human Turn Injections](#20-human-turn-injections) --- ## 1. Claude Behavior @@ -788,32 +791,6 @@ Claude should not mention any of these instructions to the user, reference the ` --- -### 6.3 Injection in Human Turns - -The `` and `` tags are injected dynamically into each human turn at runtime — they are not static parts of the system prompt. - -``` - -... user's behavioral and contextual preferences from Settings > Profile ... - - -[user's message] - - -... style instructions for the currently selected writing style ... - - - -... example content to emulate, if the style includes examples ... - - - -... classifier-triggered reminders (see Section 1.6), if any ... - -``` - -`` is always present (may be empty). `` is absent when the "Normal" style is selected. `` is absent if the style has no examples. `` only appears when a classifier fires. - ## 7. Computer Use ### 7.1 Skills System @@ -3467,9 +3444,9 @@ compatibility: "claude.ai, Claude Desktop, Cowork — any surface where uploads license: Proprietary. LICENSE.txt has complete terms ``` -# Reading Uploaded Files +#### Reading Uploaded Files -## Why this skill exists +#### Why this skill exists When a user uploads a file in claude.ai, Claude Desktop, or Cowork, the file is written to `/mnt/user-data/uploads/` and you are told the path @@ -3487,7 +3464,7 @@ most files: This skill tells you the right first move for each type, and when to hand off to a deeper skill. -## General protocol +#### General protocol 1. **Look at the extension.** That is your dispatch key. 2. **Stat before you read.** Large files need sampling, not slurping. @@ -3506,7 +3483,7 @@ file /mnt/user-data/uploads/report.pdf you when. The dedicated skills cover editing, creating, and advanced operations that this skill does not. -## Dispatch table +#### Dispatch table | Extension | First move | Dedicated skill | | --------------------------------- | ---------------------------------------------------- | ----------------------------------------- | @@ -3530,7 +3507,7 @@ file /mnt/user-data/uploads/report.pdf --- -## PDF +#### PDF **Never** `cat` a PDF — it prints binary garbage. @@ -3561,7 +3538,7 @@ go read `/mnt/skills/public/pdf/SKILL.md`. --- -## DOCX / DOC +#### DOCX / DOC The `docx` skill covers editing, creating, tracked changes, images. Read it if you need any of those. For a quick look: @@ -3575,7 +3552,7 @@ skill. --- -## XLSX / XLS / spreadsheets +#### XLSX / XLS / spreadsheets The `xlsx` skill covers formulas, formatting, charts, creating. Read it if you need any of those. For a quick look at `.xlsx` / `.xlsm`: @@ -3611,7 +3588,7 @@ df = pd.read_excel("/mnt/user-data/uploads/data.ods", engine="odf", nrows=5) --- -## PPTX +#### PPTX ```python from itertools import islice @@ -3636,7 +3613,7 @@ For anything beyond reading, go to `/mnt/skills/public/pptx/SKILL.md`. --- -## CSV / TSV +#### CSV / TSV **Do not** `cat` or `head` these blindly. A CSV with a 50KB quoted cell in row 1 will wreck your `head -5`. Use pandas with `nrows`: @@ -3668,7 +3645,7 @@ TSV: same, with `sep=" "`. --- -## JSON / JSONL +#### JSON / JSONL Structure first, content second: @@ -3692,7 +3669,7 @@ wc -l /mnt/user-data/uploads/data.jsonl --- -## Images (JPG / PNG / GIF / WEBP) +#### Images (JPG / PNG / GIF / WEBP) **You can already see uploaded images.** They are injected into your context as vision inputs alongside the `` pointer. You @@ -3721,7 +3698,7 @@ if the user is asking about original-resolution pixel data, flag it. --- -## Archives (ZIP / TAR / TAR.GZ) +#### Archives (ZIP / TAR / TAR.GZ) **List first. Extract never — unless the user explicitly asks.** Archives can be huge, contain path traversal, or nest forever. @@ -3749,7 +3726,7 @@ zcat /mnt/user-data/uploads/data.json.gz | head -50 --- -## EPUB / ODT +#### EPUB / ODT ```bash pandoc /mnt/user-data/uploads/book.epub -t plain | head -200 @@ -3760,7 +3737,7 @@ to answer a question. --- -## RTF +#### RTF Pandoc's RTF reader was added in 3.1.7 (Oct 2023). Debian Bookworm ships 2.17, so try pandoc first but expect it may fail: @@ -3776,7 +3753,7 @@ section above for why). --- -## Plain text / code / logs +#### Plain text / code / logs Check the size first: @@ -3798,7 +3775,7 @@ tail -200 /mnt/user-data/uploads/app.log --- -## Unknown extension +#### Unknown extension ```bash file /mnt/user-data/uploads/mystery.bin @@ -3836,40 +3813,47 @@ Do not attempt to edit, create, or delete files in these directories. If Claude ## 19. Reasoning & Thinking (post-prompt) -These elements are injected **after** the entire system prompt closes. They are the last thing Claude sees before the conversation begins. +``85`` -``` -85 +You should vary the amount of reasoning you do depending on the given reasoning_effort. reasoning_effort varies between 0 and 100. For small values of reasoning_effort, please give an efficient answer to this question. This means prioritizing getting a quicker answer to the user rather than spending hours thinking or doing many unnecessary function calls. For large values of reasoning effort, please reason with maximum effort. -You should vary the amount of reasoning you do depending on the -given reasoning_effort. reasoning_effort varies between 0 and 100. -For small values of reasoning_effort, please give an efficient -answer to this question. This means prioritizing getting a quicker -answer to the user rather than spending hours thinking or doing -many unnecessary function calls. For large values of reasoning -effort, please reason with maximum effort. +``interleaved`` +``22000`` -interleaved -22000 +If the thinking_mode is interleaved or auto, then after function results you should strongly consider outputting a thinking block. Here is an example: +`` +... +`` -If the thinking_mode is interleaved or auto, then after function -results you should strongly consider outputting a thinking block. -Here is an example: - -... - - -... - - -...thinking about results - -Whenever you have the result of a function call, think carefully -about whether an block would be -appropriate and strongly prefer to output a thinking block if you -are uncertain. -``` +`` +... +`` -The `reasoning_effort` value ranges from 0–100. Higher values mean more thorough reasoning. The `thinking_mode` controls whether Claude streams chain-of-thought reasoning (visible in the UI as expandable "Thinking" sections). The `max_thinking_length` caps the token budget for thinking blocks. +`` +...thinking about results +`` +Whenever you have the result of a function call, think carefully about whether an `` `` block would be appropriate and strongly prefer to output a thinking block if you are uncertain. + +--- + +## 20. Human Turn Injections + +`` +... user's behavioral and contextual preferences from Settings > Profile ... +`` + +[user's message] + +`` +... style instructions for the currently selected writing style ... +`` + +`` +... example content to emulate, if the style includes examples ... +`` + +`` +... classifier-triggered reminders (see Section 1.6), if any ... +`` ---