Apply this skill when planning, writing, restructuring, or reviewing project introductions, feature explanations, technical case studies, engineering retrospectives, build stories, launch posts, portfolio narratives, or product-education copy that must translate maker activity, implementation detail, or debugging experience into a reader's concrete situation, net benefit, credible evidence, limits, and next action.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Apply this skill when planning, writing, restructuring, or reviewing project introductions, feature explanations, technical case studies, engineering retrospectives, build stories, launch posts, portfolio narratives, or product-education copy that must translate maker activity, implementation detail, or debugging experience into a reader's concrete situation, net benefit, credible evidence, limits, and next action.
Turn product features, project work, engineering decisions, and debugging experience into useful
content organized around the reader's situation and decision. Preserve the evidence, constraints,
failed hypotheses, tradeoffs, and uncertainty that let a reader judge or reuse the result.
Use When
A project introduction, launch post, portfolio entry, feature explanation, technical case study,
engineering retrospective, incident lesson, build story, or product-education draft is created or
substantially restructured.
A draft leads with the maker's chronology, effort, stack, features, or praise and needs to lead
with the reader's problem, avoided work, avoided loss, or changed next action.
Technical metrics or capabilities need translation into workflow changes, review burden, waiting,
interruption, failure prevention, or operating cost without inventing benefits.
Development mistakes or debugging work need conversion into reusable diagnosis, decision, and
prevention guidance instead of a diary or victory story.
A reader-centered audit is requested before publishing technical or product content.
Do Not Use When
The main task is search traffic, ad-supported article structure, ranking freshness, or ad layout;
use search-ad-content-authoring first and this skill only for the reader-value layer.
The task is a repository README.md; use readme-authoring and readme-evidence-gate first.
Facts and structure are already settled and only sentence rhythm or reusable expressions need
polish; use writing-elegance.
The task is ordinary technical documentation, API reference, release notes, incident response, or
a live debugging investigation whose owning procedure is more specific.
The content would require fabricated users, quotes, measurements, failures, savings, or production
experience to become persuasive.
Required Inputs
Content surface, publishing channel, intended reader, and the decision or next action the content
should enable.
Reader situation: triggering moment, current workaround, repeated action, hidden loss, failed
attempt, switching cost, and exclusion boundary when known.
Product or project evidence: actual capability, before and after state, measurement conditions,
screenshots or examples, constraints, tradeoffs, failure cases, and verification method.
For engineering stories: reproduction conditions, rejected hypotheses and their evidence, root
cause confidence, chosen and rejected options, prevention mechanism, and diagnostic order.
Any claims that are current, numeric, regulated, security-sensitive, or externally sourced, plus
the owning verification or freshness procedure.
Relevant command-intent entries when repository files change.
Preconditions
The task matches the Use When conditions and does not match the Do Not Use When exclusions.
The reader, surface, and intended decision are specific enough to distinguish a useful scene from
a generic persona such as "busy professionals" or "all creators."
Claims can be tied to supplied or repository-supported evidence, or explicitly marked as
hypotheses, examples, or unknowns.
Allowed Edits
Create or revise titles, openings, outlines, paragraphs, evidence blocks, comparisons, limitations,
calls to action, and channel-specific variants for the selected content.
Reorder maker chronology into reader decision order and cut implementation detail that does not
help the reader identify, trust, evaluate, or act on the result.
Preserve technical identifiers, measurements, commands, and factual boundaries unless the owning
skill and evidence support changing them.
Do not invent ROI, time saved, accuracy, user behavior, operational impact, root cause, or customer
quotes. Do not convert a possible benefit into a universal promise.
Do not hide setup, migration, learning, review, privacy, money, reliability, or organizational cost
merely to improve conversion.
Procedure
Name the reader's decision. State what the reader should be able to recognize, compare, try,
avoid, or decide after reading. If no decision exists, narrow the content before polishing it.
Build a reader-situation ledger:
triggering moment and visible object or action;
current workaround or failed attempt;
repeated nuisance, waiting, interruption, rework, risk, embarrassment, or avoided loss;
transition cost and the next action after receiving value;
people for whom the offer or lesson does not fit.
Build an evidence ledger. Separate observed facts, measured comparisons, reproduced causes,
plausible inferences, and unknowns. Keep each important claim next to the evidence and limitation
that lets a skeptical reader judge it.
Translate capability into net benefit through a complete chain:
capability -> removed or reduced action -> changed workflow -> avoided cost or enabled outcome -> next action.
Stop at the last supported link. Subtract setup, learning, migration, waiting, review, error, and
switching cost rather than presenting gross benefit as net benefit.
Translate technical measures into reader-verifiable consequences. Explain latency as avoided
waiting or context switching, accuracy as review or rework remaining, automation as the points
where a person still intervenes, integration as transfers or logins removed, reliability as what
survives failure, and security as access, traceability, or loss containment. Keep the original
metric and measurement conditions when they matter.
Choose an opening scene, not a demographic label. Prefer a specific moment, workaround, object,
quiet failure signal, contradictory state, or familiar inner sentence. Do not manufacture shame,
fear, or fake urgency.
Order the content by reader questions: Is this my problem?, What changes?, How does it work?,
Why should I believe it?, What will it cost or fail at?, and What can I do next?. Maker
chronology may appear later as supporting evidence.
For project and feature content, show the result before explaining the build. Use a real example,
before-and-after artifact, screenshot, demo, or bounded measurement when available. Describe the
strongest tradeoff instead of stacking generic adjectives.
For engineering and debugging stories, write the diagnostic value:
failure conditions and smallest useful reproduction before the final patch;
the mistaken assumption or hypothesis, the observation that rejected it, and the evidence for
the accepted cause;
rejected options and the condition that made the chosen option less harmful;
before and after measurements under the same conditions;
prevention through tests, validation, observability, or workflow changes;
the first checks the next person should perform.
Distinguish disappearance from addition. Prefer the work, wait, risk, or workaround that no
longer remains over a list of new features. For AI-assisted outcomes, state the human review,
provenance, editability, and worst ordinary result needed for the reader to trust and own the
output.
Make competence inferable. Show constraints, decisions, verification, rejected alternatives,
maintenance reduction, and the boundary between the author's contribution and tools or prior
work. Do not use effort, code volume, development time, praise, or borrowed authority as proof.
Adapt the packaging to the channel. A project name may belong after the reader outcome in an
unfamiliar audience, while a known repository or release surface may require the exact name
first. Create title, opening, image, and excerpt variants when audiences have materially
different reasons to care; do not change the factual promise between variants.
End with one small next action or a concrete changed future action. Do not end with maker
gratitude, vague ambition, or simultaneous demands to buy, subscribe, comment, and share.
Run the reader audit:
Does the first screen contain a recognizable situation and supported benefit?
Would the piece still matter with the maker and project names hidden?
Does every paragraph answer a reader question?
Are evidence and limitations adjacent to the claims they qualify?
Are costs, exclusions, worst ordinary behavior, and human review visible?
Can the reader state the next action without guessing?
Run the narrowest configured verification covering changed docs, content, templates, package
surfaces, and mustflow contracts.
Postconditions
The content begins from a recognizable reader situation or decision rather than maker effort.
Benefits are expressed as supported net workflow change, avoided loss, or enabled next action.
Technical stories preserve reproduction, hypothesis rejection, cause confidence, tradeoffs,
comparable evidence, prevention, and a reusable diagnostic order when those facts exist.
Costs, exclusions, uncertainty, contribution boundaries, and verification remain visible.
Search, README, documentation, release, incident, and style-only responsibilities stay with their
owning skills.
Verification
Use configured oneshot command intents when available:
changes_status
changes_diff_summary
docs_validate_fast
test_release
mustflow_check
Use the owning content, docs, source-freshness, numeric, security, or repository verification intent
when the claims depend on those surfaces.
Failure Handling
If the reader or next decision is unknown, return the missing decision boundary instead of writing
generic copy for everyone.
If benefit evidence is absent, describe the capability and testable hypothesis without inventing
time, money, accuracy, adoption, or emotional outcomes.
If the root cause was not established, label the workaround and unresolved cause separately; do
not publish accidental recovery as diagnosis.
If before and after conditions differ, report the measurements separately and remove the causal or
percentage comparison.
If a limitation makes the claimed benefit negative for the target reader, narrow the audience,
change the claim, or reject the content premise.
If channel optimization would hide costs, distort evidence, or alter the promise, preserve the
factual contract and report the packaging tradeoff.
If a current or high-stakes claim cannot be verified, omit, qualify, or route it through the
relevant freshness, numeric, security, legal, or domain procedure.
Output Format
Content surface, channel, reader situation, and intended decision
Current workaround, removed work, net benefit, transition cost, exclusion boundary, and next action
Evidence ledger, limitations, tradeoff, contribution boundary, and claim changes
Engineering-story reproduction, rejected hypotheses, cause confidence, prevention, and diagnostic
order when applicable
Title, opening, outline, evidence placement, ending, and channel variants changed
Files changed
Command intents run and skipped checks
Remaining reader-value, evidence, or publication risk
