| name | accelint-english-manager |
| description | Use when the user wants prose rewritten, tightened, audited, simplified, polished, humanized, grammar-checked, or made plainer, clearer, shorter, easier to scan, or easier to act on without changing the intended meaning, audience, tone, or explicit constraints. Trigger on requests such as plain English, simple English, make this clearer, make this more direct, clean this up, edit this, review this writing, grammar check, too wordy, too formal, less fluffy, friendlier, shorter, audit then rewrite, keep the tone, mode=strict, STE, ASD-STE100, or ADHD-friendly. Also trigger for docs, prompts, emails, UI copy, support replies, release notes, status updates, incident notes, procedural text, and other LLM-written prose. Prefer this skill when the main job is improving English prose or preserving tone while increasing clarity, including audit-only requests and exact-text-preservation constraints, but not when the real task is fact-checking, policy setting, or substantive content design. |
| license | Apache-2.0 |
| metadata | {"author":"accelint","version":"1.3.9"} |
English Manager
Use plain, direct English that is easy to read, easy to scan, and easy to act on.
Do not optimize for brevity alone. Preserve the user's intended meaning, audience, tone, and explicit constraints while making the text clearer, steadier, and harder to misread.
This skill uses one default writing system:
- plain-language discipline for direct, concrete wording
- STE-leaning structure for technical clarity and stable terminology
- ADHD-friendly shaping for scanability and actionability when the text helps someone do something
Use these together. They are not separate modes.
Hard constraints
These outrank style preferences.
- Preserve the user's meaning before you optimize wording.
- Preserve the requested tone, audience fit, and explicit format constraints.
- Preserve deliberate warmth, rhythm, humor, persuasion, or brand voice when the user wants them.
- Preserve code, identifiers, commands, file paths, quoted errors, product names, API names, config keys, and legal text unless the user explicitly asks to rewrite them.
- Keep real uncertainty, real nuance, and real obligation levels. Do not make text sound simpler by making it less true.
- Do not claim official ASD-STE100 compliance. If the user asks for strict STE, say that full compliance depends on the official standard and dictionary.
Start here
Choose the smallest fitting path before you edit.
Step 1: Ask for the mode first
Do this first for drafting or rewriting tasks unless the user already specified the mode explicitly.
Offer these choices:
mode=default — local rewrite by default, plain and directmode=strict — stricter technical control, structural rewrite allowed when needed
If the user did not choose a mode, ask a short clarifying question instead of assuming.
Done when: the rewrite mode is explicit, or the request is clearly audit-only.
Step 2: Choose the output mode
Requires: Step 1 is complete when the request needs a rewrite mode.
Choose one output mode before you edit:
- Audit only — review the text without rewriting it unless the user asks.
- Rewrite only — return cleaner final text directly.
- Audit plus rewrite — give findings first, then the rewrite.
If the user asks for only the rewrite, return only the rewrite. Do not return audit notes unless the user asked for them.
Done when: the output mode is explicit.
Step 3: Preserve the constraints
Requires: Step 2 is complete.
Do this before you rewrite or restructure anything.
Extract and protect these first:
- meaning
- audience
- tone
- format
- exact wording that must stay
- quoted or technical text that must stay exact
If a style rule conflicts with an explicit user constraint, the constraint wins.
Done when: the protected constraints are identified.
Step 4: Choose the scope through the mode
Requires: Step 3 is complete.
Let the mode set your default rewrite scope.
mode=default — do a local rewrite by default. Tighten or clarify the given text only. Preserve the source structure unless the user clearly asks for a fuller artifact or the current structure hides key meaning.mode=strict — allow a structural rewrite when stronger control helps the text do its job. Reorganize, split, or relabel content when the current structure hides sequence, logic, safety, requirements, or operational clarity.
Do not treat mode=strict as permission to broaden the task casually. In both modes, keep the smallest change that solves the real problem unless the user explicitly asks for a broader rewrite.
Done when: the allowed rewrite scope is clear.
Default writing method
Use this method unless the task clearly calls for a stricter overlay.
1. State the point early
Lead with the point, result, or next action.
- In explanatory text, put the main fact early.
- In operational text, put the next action early.
- In support or error help, state what happened, why if known, and what to do next.
Do not add a preamble when the answer works better without one.
2. Use direct, concrete wording
Write the clearest accurate sentence you can.
- prefer concrete actions and results over abstraction
- use short familiar words when they keep the same meaning
- remove filler, recap padding, and inflated phrasing
- name the actor when the actor matters
- prefer active voice when it improves clarity
If a simpler word would reduce accuracy, keep the precise word.
3. Keep one term for one concept
Pick one term for each repeated concept and keep it stable.
Do not rotate synonyms just to avoid repetition.
4. Keep the text easy to scan
Use light ADHD-friendly shaping by default:
- keep sentences and paragraphs bounded
- keep one main action or fact per sentence when possible
- use lists only when they make the work easier to follow
- keep side issues out of the main path until they matter
- make the next action obvious when the reader needs to act
Do not force checklist structure onto casual, creative, or voice-sensitive prose.
5. Match the structure to the job
Use the shape that fits the writing.
- Procedural or operational writing: imperative steps, condition before command, one action per step, warnings as explicit instructions
- Descriptive writing: one main fact per sentence, related facts grouped together, no buried instructions in explanation paragraphs
- Voice-sensitive or persuasive writing: keep the clarity gains without flattening the voice that does the job
6. Keep necessary nuance
Write plainly without overstating certainty or force.
- keep real hedges when the fact is uncertain, conditional, estimated, or incomplete
- remove fake hedges when they only add fog
- preserve requirement strength, recommendation strength, permission, and capability accurately
- keep passive voice only when the actor is unknown, irrelevant, tact-sensitive, or deliberately de-emphasized
Mode control
Use these writing modes when they materially help.
| Mode | When | Apply |
|---|
| mode=default | Most requests | Plain-language discipline, stable terminology, scanable structure, action-first shaping when useful, and local rewrites by default |
| mode=strict | The user explicitly asks for STE, ASD-STE100-style writing, very strict plain language, highly controlled technical wording, or a stricter audit of technical prose | Default mode + stronger STE structure, stronger modality discipline, stronger procedural/descriptive separation, tighter terminology control, and structural rewrites when needed for clarity or control |
If the user asks for strict STE-style review:
- Say that you can do a strict STE-leaning review.
- State that official ASD-STE100 compliance depends on the official standard and dictionary.
- Load only the relevant part of
references/ste-rules.md that the request needs.
- Cite rule numbers only from the part of
references/ste-rules.md that you actually loaded.
- If you did not load the relevant rule text, do not cite rule numbers.
Do not combine these actions into one sentence. Keep the order explicit.
If the user asks for "plain English," "simple English," "clean this up," or a similar generic cleanup request without naming a mode, treat that as a plain-language goal, not as implicit mode=strict.
Serial-order detection and handling
Treat sequence as behavior when the text tells the reader or agent to do one thing before another.
Detect and strengthen these cases:
- explicit serial instructions — numbered steps,
first/next/then/finally, before/after/until, requires, depends on, done when, return to
- implied step ordering — one sentence hides multiple actions, one action depends on the output of another, or a warning implies a missing gate
- sequencing cues in skill files — ask-for-mode first, choose output mode before delivery, preserve constraints before rewriting, load rule references before citing them, run self-check before delivery
- sequencing cues in general prose — procedures, operational notes, support steps, onboarding text, prompts, and any paragraph where setup, action, validation, and delivery are mixed together
When order matters:
- make the sequence explicit
- use a numbered list for 2 to 3 short ordered steps
- use
### Step 0 plus a checklist, then ### Step N: Name, for 4 or more ordered steps
- add
Requires: when a step depends on an earlier step
- add
Done when: when later work depends on a successful check
- add a named failure route when a later step must wait for a passing check
- split
do X and then Y into separate steps when that improves compliance
- put conditions before actions when that makes timing clearer
- replace emphasis-only warnings with enforceable gates only when a later action already depends on an earlier check
- never leave ordered work in plain bullets
Load references/serial-instruction-guidance.md before you rewrite or audit any text with real workflow order, gating, branching, or validation loops. Keep SKILL.md operational. Use the reference for deeper detection logic, structure rules, and edge cases.
Output rules
Audit only
Default to a compact review unless the user asks for something more formal.
Use this shape:
- Summary — 1 to 3 sentences on the main clarity, tone, or actionability issues
- Highest-risk issues first — especially meaning drift, ambiguity, hidden actions, obligation drift, or broken structure
- Targeted findings — source text, risk, and a brief note only when it helps
- Optional full rewrite — include it only if the user asked for one or the passage has repeated issues
Rewrite only
Return only the rewrite when the user asks for only the rewrite.
Do not prepend audit notes or explanation unless the user asked for them.
Audit plus rewrite
Give the findings first, then the rewrite.
Reference loading map
Load references only when they materially help.
references/serial-instruction-guidance.md — ordered instructions, implied sequencing, gating, branching, validation loops, and when a paragraph should become explicit stepsreferences/substitutions.md — wording cleanup, filler removal, consistency sets, modality checksreferences/checklist.md — final verification pass or detailed audit supportreferences/ste-rules.md — technical, procedural, instructional, or explicit STE requests; also for mode=strictreferences/adhd-patterns.md — stronger action-oriented shaping when the reader needs execution help, low-friction next steps, or ADHD-friendly formatting beyond the defaultreferences/use-cases.md — adapt the core method to docs, prompts, support replies, incident notes, UI copy, or voice-sensitive writingreferences/rfc-2119.md — only for normative or behavior-defining text such as policy, requirements, interface contracts, governance, or procedures where informal severity labels hide true obligationreferences/examples.md — anchor patterns when the request is ambiguous or the text mixes clarity goals with voice-sensitive constraints
Delivery workflow
For drafting from scratch
Step 1: Ask for the mode
Do this first unless the user already specified the mode.
Done when: the rewrite mode is explicit, or the request is truly audit-only.
Step 2: Choose the output mode
Requires: Step 1 is complete when the request needs a rewrite mode.
Choose Audit only, Rewrite only, or Audit plus rewrite before you draft.
Done when: the output mode is explicit.
Step 3: Extract the constraints
Requires: Steps 1 and 2 are complete.
Identify the meaning, audience, tone, format, and exact wording that must stay fixed before you draft.
Done when: the protected constraints are identified.
Step 4: Apply the mode
Requires: Step 3 is complete.
Let mode=default or mode=strict set the default scope.
Done when: the allowed rewrite scope is clear.
Step 5: Draft in direct English
Requires: Steps 1 to 4 are complete.
Write in plain, direct English that preserves the user's intended meaning, audience, tone, and explicit constraints.
Done when: the draft does the same job as the source.
Step 6: Keep repeated terms stable
Requires: Step 5 is complete.
Use one term for one concept. Do not rotate synonyms for style.
Done when: repeated concepts use stable terms.
Step 7: Make the action path easy to scan
Requires: Steps 5 and 6 are complete.
Do this when the reader must act.
If the text contains ordered actions, load references/serial-instruction-guidance.md before you make the order explicit.
Done when: ordered actions are visible as ordered actions.
Step 8: Run the self-check before delivery
Requires: Steps 1 to 7 are complete.
Do not deliver before Step 8 is complete.
If the self-check fails, return to the earliest affected step.
Done when: the self-check passes.
For revising existing text
Step 0: Track progress
Do this before any other work when the rewrite workflow has 4 or more real actions.
Create a short checklist in your working state or reply and update it after each step.
Step 1: Ask for the mode
Do this first unless the user already specified the mode.
Done when: the rewrite mode is explicit, or the request is truly audit-only.
Step 2: Preserve the user's meaning and constraints
Requires: Step 1 is complete.
Do this before any rewrite.
Done when: meaning, audience, tone, format, and untouchables are identified.
Step 3: Apply the mode
Requires: Steps 1 and 2 are complete.
Let mode=default or mode=strict set the default scope.
Done when: the allowed rewrite scope is clear.
Step 4: Tighten the wording with the smallest safe change
Requires: Steps 1 to 3 are complete.
Do one safe rewrite pass that preserves the source job, meaning, tone, audience, and explicit constraints.
Done when: the draft is clearer without changing the job.
Step 5: Remove filler, stale phrasing, and avoidable abstraction
Requires: Step 4 is complete.
Remove wording that adds fog without adding meaning.
Done when: the remaining wording is carrying meaning or control.
Step 6: Split overloaded sentences or restructure only when substitution alone will not work
Requires: Steps 4 and 5 are complete.
If the source hides sequence, gates, or branch logic, load references/serial-instruction-guidance.md before you rewrite the sequence explicitly.
If sequence checks fail, return to Step 4.
Done when: the workflow cannot be read as unordered advice.
Step 7: Recheck pronouns, modality, and untouchables
Requires: Steps 4 to 6 are complete.
Done when: referents are clear and obligation strength still matches the source.
Step 8: Run the self-check before delivery
Requires: Steps 1 to 7 are complete.
Do not deliver before Step 8 is complete.
If the self-check fails, return to the earliest affected step.
Done when: the self-check passes.
For checking text instead of rewriting it
Step 1: Identify the highest-risk issues first
Put meaning drift, ambiguity, hidden actions, obligation drift, and weak sequencing first.
Done when: the highest-risk issues are ranked.
Step 2: Keep the review proportional to the request
Requires: Step 1 is complete.
Done when: the audit depth matches the request.
Step 3: Use a stricter, rule-based audit format only when the user asks for it or the text is high-consequence
Requires: Steps 1 and 2 are complete.
Done when: the audit shape matches the risk and request.
Step 4: Load the right references before citing them
Requires: Step 3 is complete.
If the user asked for STE checking, load only the relevant part of references/ste-rules.md that you need. Do not invent rule numbers.
If the text contains ordered actions or weak sequencing cues, load references/serial-instruction-guidance.md before you judge whether the sequence is clear enough.
Done when: every cited rule comes from a loaded reference.
Step 5: Run the self-check before delivery
Requires: Steps 1 to 4 are complete.
Do not deliver before Step 5 is complete.
Done when: the self-check passes.
Conflict resolution
Use this priority order when rules compete:
- User meaning, audience, tone, and explicit constraints
- Untouchables and required exact wording
- Clarity and actionability
- Technical-writing discipline
- Brevity
Specific resolutions:
- preserve warmth or voice when the text depends on relationship tone, persuasion, rhythm, or brand fit
- use STE-like discipline as a clarity tool, not as a form override for creative or hybrid writing
- remove fake hedges, but keep real uncertainty
- use active voice by default, but keep passive when tact, uncertainty, or deliberate emphasis matters
Required self-check before delivery
Before you deliver, confirm:
- The rewrite still does the same job for the same audience.
- Key terms stayed stable.
- Obligation, permission, capability, and uncertainty did not drift.
- Untouchables stayed exact.
- Any explicit or implied ordered instructions are still visible as ordered instructions.
- Any required gate, dependency, or validation step still blocks the later step clearly.
- The final answer matches the requested output mode.
- The final answer matches the requested rewrite mode.
- Any remaining qualitative wording still has a clear operational role and does not act as a hidden gate.
For a deeper mechanical pass, load references/checklist.md.
Limits
This skill improves clarity and execution-readiness. It does not replace subject-matter accuracy, legal review, brand review, or official ASD-STE100 certification.
It is also a writing skill, not a fact-checking or policy-setting skill. Improve the prose without inventing new requirements, commitments, or product behavior.
When you apply this skill, optimize for disciplined English that stays precise, truthful, easy to scan, and easy to act on.
