affirm

name	affirm
description	Rewrite negative instructions as positive directives and remove hedging language. Use when user invokes /affirm or asks to make instructions direct, remove hedging, flip negatives to positives, or strengthen language in skill files, docs, or prompts. Works on inline text, files, or conversation context. Target instruction and directive text only; route general prose editing to /distill.

Affirm

Rewrite instructions as direct positive statements. Remove hedging.

Parameters

• text (required): The text to rewrite. Inline text, file path, or conversation context.

text = <args> or conversation context
if no text provided: AskUserQuestion("What text should I rewrite?")
if text looks like a file path (starts with / or ./, ends with a known extension, or matches an existing file):
  read the file at that path  // Read file content
  apply all transforms to the file content
  write the result back to the same path  // Write result back
  show a summary of changes made

// Flip negatives
do("find sentences with 'do not', 'don't', 'never', 'avoid', 'no', 'not'")
do("rewrite each to state the desired behavior directly")

// Collapse double negatives
do("find double negatives: 'not uncommon', 'not unlikely', 'not insignificant', 'not without'")
do("replace with the direct positive: 'common', 'likely', 'significant', 'with'")

// Remove hedging
do("find hedging words: 'might', 'perhaps', 'consider', 'try to', 'should probably', 'it seems', 'be careful to', 'make sure to', 'ensure that'")
do("replace each with direct instructions")

// Strip permission framing
do("find permission patterns: 'you can', 'you may', 'feel free to', 'you are allowed to'")
do("replace with imperative: 'use X' instead of 'you can use X'")

// Remove vacuous conditionals
do("find 'if applicable', 'when appropriate', 'as needed', 'where possible'")
do("if the condition is always true in context, remove it")
do("if the condition is genuinely conditional, leave it")

// Verify
do("check that meaning is preserved — positive framing changes phrasing, not intent")
do("check that scale descriptions and genuine uncertainty markers are left intact")

// Output
do("show before/after for each changed sentence")

Reference

Negative to positive

The model reads "do X" in one inference step. "Do not do Y" requires two: parse Y, then invert. Positive framing resolves in one step and leaves less room for misinterpretation.

Before	After
Do not use mocks in integration tests	Use real database connections in integration tests
Never commit secrets	Keep secrets out of version control
Avoid long functions	Split functions at natural boundaries
Don't repeat yourself	Extract shared logic into a single location

Keep negatives when the positive form loses precision. "This function returns null on failure" says more than "This function returns a value on success."

Hedging to direct

Hedging signals optionality. The model reads "consider X" as "X is one possibility among many" and may skip it. Direct instructions close that gap.

Before	After
You might want to check the logs	Check the logs
Consider using a cache here	Use a cache here
Try to keep functions short	Keep functions short
Make sure to validate input	Validate input
Be careful not to break the API	Preserve the API contract
Be honest about uncertainty	Calibrate grades to match actual confidence

Preserve hedging in scale descriptions and genuine conditional uncertainty ("if the data is unavailable, the result may be incomplete"). These describe reality, not instructions.

Double negatives

Two negations require two inversions to reach the meaning. Replace with the direct positive.

Before	After
not uncommon	common
not unlikely to fail	likely to fail
not without risk	risky
not insignificant	significant

Permission framing to imperative

In instruction context, "you can" signals optionality. The model treats optional steps as skippable. Use imperative form.

Before	After
You can use SQLite for local caching	Use SQLite for local caching
Feel free to split the function	Split the function
You may want to add a timeout	Add a timeout

Preserve permission framing in user-facing text where the user genuinely has a choice ("You can override this with --force").

Vacuous conditionals

"If applicable" and "when appropriate" create an escape hatch. When the condition is always true in context, the qualifier adds nothing. Remove it.

Before	After
Validate input, if applicable	Validate input
Use caching where possible	Use caching
Add error handling as needed	Add error handling

Preserve genuinely conditional qualifiers where the condition is sometimes false ("if the user provides a file path, read it").