Brand Voice & Tone Rewriting Skill
Overview
Given a draft piece of workplace writing and a tone target, the agent must produce a structured four-part response: explicit tone anchors, a severity-tagged audit of deviations, a rewritten draft with inline change annotations, and a per-anchor compliance summary — escalating to two alternative versions whenever a conflict between the writer's intent and the target tone is detected.
Your submission is a single SKILL.md file that teaches a fixed base agent how to perform brand voice and tone rewriting across varied inputs: style guides, exemplar messages, vibe cues, or combinations of these. The agent must apply the same disciplined procedure regardless of the writing format (Slack message, Notion section, status update, announcement, release note, support reply) or how the tone target is expressed.
Strong skills teach the agent a transferable reasoning procedure: how to derive concrete anchors from whatever tone evidence is available, how to audit a draft systematically before touching a word, how to annotate every rewriting decision, and how to recognise when the right move is to hold back — returning input unchanged, asking a single clarifying question, or surfacing a tension rather than silently resolving it.
Weak skills hand the agent a template or a set of memorised rewrites. These break the moment surface details change — a new domain, a differently phrased style guide, or a draft where the correct answer is to do nothing.
A skill built to memorise specific scenarios will not hold up as challenges evolve.
You submit a SKILL.md file to the challenge. The validator attaches it to our canonical base agent, runs it against all 10 scenarios, and returns a scored breakdown. Top submissions earn Alpha emissions daily and on displacement.
We are looking for skills that demonstrate good design and help the base agent behave like a capable practitioner in this domain.
Requirements (10 Tests)
Your skill is tested across 6 capability types, with 10 scenarios in total. Every submission is evaluated against all 10 scenarios — there is no sampling, so consistent quality across every capability type counts.
| Category | Overview | Test cases |
|---|---|---|
| Type A — Intent-vs-tone conflict detection | Tests whether the agent can identify situations where making a draft tone-compliant would distort, soften, or obscure the writer's actual intent. The skill must teach the agent to name the tension explicitly and present two clearly labelled alternatives — one tone-compliant, one preserving literal intent — rather than silently laundering the message into style conformance. | 1 |
| Type B — Near-compliant minimal rewriting | Tests whether the agent exercises proportionate restraint when a draft is mostly on-tone. The skill must teach the agent to audit the draft thoroughly, tag only genuine deviations by severity, and limit rewrites to those deviations — rather than over-editing a draft that is already largely compliant. | 1 |
| Type C — No-deviation recognition | Tests whether the agent can correctly identify a draft that has zero material deviations from the tone target and return it unchanged. The skill must teach the agent that rewriting for its own sake is a failure mode, and that a well-reasoned 'no changes needed' conclusion — supported by a completed compliance summary — is a valid and complete response. | 1 |
| Type D — Register-shift rewriting | Tests the agent's ability to rewrite across a significant register gap — for example from casual or rambling prose to structured technical writing, or from hedged corporate language to direct executive voice. The skill must teach the agent to anchor every change to a specific audit finding or tone anchor and surface that reasoning inline via bracketed annotations. | 1 |
| Type E — Standard full rewrite | Tests end-to-end execution of the full four-part procedure on a draft with multiple clear deviations from the tone target. The skill must teach the agent to move through anchor derivation, severity-tagged auditing, annotated rewriting, and per-anchor compliance scoring in order — preserving all semantic content and flagging any length-budget implications. | 1 |
| Type F — Subtle intent tension | Tests whether the agent can detect low-signal intent-vs-tone conflicts where the tension is not obvious — for example where a hedged phrase carries deliberate meaning the writer wants preserved, or where a tone target's 'no hedging' rule would strip a genuine caveat. The skill must teach the agent to treat subtle ambiguity as a reason to surface and name the tension, not to resolve it silently. | 1 |
Surface details — names, organisations, word counts, domain specifics, and the exact phrasing of tone targets — rotate between variants while the underlying capability being tested remains the same.
Datasets
Available Datasets
Your skill attaches to a standardised base agent provided by sundae_bar. The base agent is the same for every submission. It is lean by design: generic memory blocks, standard tools, no customer-specific knowledge. Its performance on the evaluation scenarios is driven almost entirely by the skill you give it.
You do not submit or modify the base agent. Its behaviour is fixed across all submissions so that scoring reflects skill quality, not agent configuration.
Tools available to the base agent: the load_skill tool (which loads your SKILL.md content when the agent decides to use it), plus standard built-ins including memory and conversation search. Available tools: load_skill.
Important: the validator attaches your skill to the base agent at evaluation time. You cannot add tools, modify memory blocks, or alter the base agent configuration.
Tip: write your skill as though you are onboarding a capable new hire who needs a procedure, not scripting a chatbot.
Evaluation Criteria
Each scenario is graded by an LLM judge. Final scores combine two graders and a set of deterministic penalties.
| Criteria | Weight | |
|---|---|---|
Skill use An LLM judge evaluates whether the agent's output actually followed the procedures your skill teaches, rather than freelancing or pretending to use the skill. | 50 % | |
Scenario quality An LLM judge evaluates how well each output meets the per-type quality bar, independent of whether your skill was followed. A reference good output anchors the judge's sense of the quality bar. | 50 % |
Each scenario is scored on the quality of the agent's response; deterministic penalties apply when the agent makes moves that damage output quality — such as rewriting a compliant draft unnecessarily, silently dropping semantic content, or failing to surface a detectable intent-vs-tone conflict. Scenario scores aggregate to capability-type averages, which roll up to the final challenge score shown on the leaderboard.
Submission
How to submit
Create an account and add your Bittensor wallet ID to your profile to earn alpha rewards in the challenge.
Review the reference skills (good, mediocre, gaming) to understand the quality bar and calibration points.
Draft your SKILL.md. Focus on teaching procedure rather than providing templates.
Validate locally using the published dataset variants to sanity-check output quality.
Submit through the sundae_bar challenge portal by uploading your SKILL.md file.
Review your scored breakdown and resubmit.
Constraints
Submission format: a single SKILL.md file with YAML frontmatter (name, description) and Markdown body.
Base agent: fixed. Provided by sundae_bar.
Runtime: the validator attaches your skill to the base agent and runs all 10 scenarios per submission.
Tests per submission: 10 (the full scenario set — no sampling).
Submission limits: 3 submissions per day. A rolling challenge limit also applies.
Rewards
Emissions accrue on-chain to whichever submission currently holds the top of the leaderboard. There is no fixed per-submission payout: the current leader earns for as long as it stays first.
Time at the top takes all. A strong skill submitted early starts accruing immediately and keeps earning until another submission displaces it. Once you are displaced, accrual stops until you reclaim the top.
Your SKILL.md is scanned at submission time by a skill integrity gate. The gate checks that your skill teaches procedure rather than encoding scenario-specific content from our evaluation pool.
A skill that fails the integrity gate scores 0 for the submission. The gate runs deterministically and reports its finding as part of your results. This is how the evaluation framework stays credible: a high score should reflect a skill that genuinely teaches procedure, not one that memorised our test set.