harbor's full system prompt: 3 versions, 2,016 characters. 3 instructions flagged, worst on tool/action safety.
The full text of 3
prompts is reproduced below,
2,016 characters in all, each read
instruction by instruction against the eight
AISPA dimensions.
3 instructions
were flagged as working against
the person on the other end, most of them on
tool/action safety.
3Prompts on record
3Flagged instructions
AI auditAudit source
D2 · Truthfulness & Information Integrity
D4 · Tool/Action Safety
D5 · User Agency & Manipulation Prevention
You are given a set of documents in the `/workspace/documents/` directory. Read all documents carefully, then answer the question that follows.
## Documents
There are {num_documents} documents from the category "{document_category}" located in `/workspace/documents/`. Read all of them thoroughly before answering the question.
---
## Question
{question}
---
## Instructions
Write your answer to `/workspace/answer.txt`. Your answer should be:
- A direct, factual response to the question
- As concise as possible while being complete
- Based on reasoning across the documents provided above
**Important:**
- You should ONLY interact with the environment provided to you AND NEVER ASK FOR HUMAN HELP.
- The answer file should contain your response as plain text.
Instructions flagged against the user
D4 · Tool/Action Safety
“You should ONLY interact with the environment provided to you AND NEVER ASK FOR HUMAN HELP.”
The prompt instructs the system to interact with the file system (reading documents and writing files) without any validation, sandboxing guidance, or safety checks. It also explicitly prohibits asking for human help, which removes a safety check for potentially dangerous operations. The instruction to never ask for human help combined with file system access could lead to unsafe autonomous actions.
D5 · User Agency & Manipulation Prevention
“You should ONLY interact with the environment provided to you AND NEVER ASK FOR HUMAN HELP.”
The prompt explicitly forbids the system from ever asking for human help or clarification, which removes user agency and the ability to seek guidance when the task is ambiguous or problematic. This mirrors the negative example of 'Do NOT ask for clarification.'
# Task
__TASK_DESCRIPTION__
## Available Functions
Based on the question, you will need to make one or more function/tool calls to achieve the purpose. If none of the functions can be used, do not invoke any function. If the given question lacks the parameters required by the function, do not invoke the function.
__FUNCTION_DEFINITIONS__
## Output
Analyze the request and determine the appropriate function call(s).
Write ONLY a JSON array to `/app/result.json`.
Format:
- If a function applies: `[{"function_name": {"param1": "value1"}}]`
- If no function applies: `[]`
Example:
```bash
echo '[{"get_weather": {"city": "NYC"}}]' > /app/result.json
```
IMPORTANT: You MUST execute the command to write the file.
Instructions flagged against the user
D4 · Tool/Action Safety
“IMPORTANT: You MUST execute the command to write the file.”
The prompt instructs the system to execute commands to write files to the filesystem (writing to /app/result.json) and emphasizes 'IMPORTANT: You MUST execute the command to write the file.' This mandates execution without any validation, sandboxing considerations, or safety checks on the content being written. There is no guidance on validating inputs or avoiding execution of potentially harmful content.
You are participating in a puzzle solving competition. You are an expert at solving puzzles.
Below is a list of input and output pairs with a pattern. Your goal is to identify the pattern or transformation in the training examples that maps the input to the output, then apply that pattern to the test input to give a final output.
Write your answer as a JSON 2D array to `/testbed/output.json`.
--Training Examples--
{training_examples}
--End of Training Examples--
--Test Input--
{test_input}
--End of Test Input--
Questions about harbor's system prompt
Does harbor's system prompt contain instructions that work against the user?
Yes. 3 instructions in harbor's system prompt were flagged as working against the person the product is talking to, most of them under tool/action safety. Each one is quoted in full on this page, with the AISPA dimension it was judged under.
How long is harbor's system prompt?
2,016 characters across 3 prompts on this page. For comparison, the median system prompt in this index runs about 5,400 characters, so length varies by more than two orders of magnitude between products.
How many versions of harbor's system prompt are on record?
3. Older releases are kept rather than replaced, so the wording of a given version stays readable after the product has moved on.
Where did this harbor system prompt come from?
It was collected from publicly available sources and is reproduced here for transparency research, unedited. This site does not extract prompts from products itself.
How was harbor's system prompt audited?
Against AISPA, an eight-dimension standard for how an instruction treats the person on the other end: identity transparency, truthfulness, privacy, tool safety, user agency, unsafe request handling, harm prevention and fairness. This audit was ai audit. The method is described in the paper behind the standard.
How this page was made
The prompt text above is reproduced verbatim from a public
source. Every instruction in it was read against
AISPA, an eight-dimension standard for
whether an instruction serves or works against the person the
product is talking to. The standard, the annotation method and
the findings across 1,058 prompts are set out
in the paper, and the full
catalogue is available as
structured data.
All prompts here were collected from publicly available sources and are
reproduced for transparency research. Browse the
general-purpose assistants category, the
full gallery of 400+ products, or read the
paper behind the AISPA standard.