← Compound WritingCONTENT HISTORYWHAT CHANGED · RULE-BASED ANALYSIS
Update to Compound Writing
Snapshot Sep 30, 2026 · 23:16 UTC · version 2.4.1
Collection source: not recorded for this historical snapshot.
First saved snapshot
No earlier snapshot is available to establish a change.
Compare saved observations
Download comparison JSONFull technical diff · 0 changed fields
Full snapshot data
{
"description": "Capture a confirmed writing preference, editorial lesson, recurring failure pattern, or workflow improvement in the right durable context file. Use when the user asks to save, remember, codify, or turn a repeated correction into a rule.",
"included_files": [],
"name": "cw-save",
"skill_md_contents": "---\nname: cw-save\ndescription: Capture a confirmed writing preference, editorial lesson, recurring failure pattern, or workflow improvement in the right durable context file. Use when the user asks to save, remember, codify, or turn a repeated correction into a rule.\n---\n\n# Save\n\nTurn a confirmed observation into maintained context so future work can use it. Do not claim memory unless a real file or system is updated.\n\n## Load The Context Contract\n\nRead `../../references/context-contract.md` and inspect the active project's instruction and style files before choosing a destination.\n\n## Classify The Learning\n\n| Learning | Preferred destination |\n|---|---|\n| Syntax, diction, tone, cadence, register, punctuation, verbal tic, or sentence-construction preference | The governing `VOICE.md` |\n| Argument, evidence, article structure, substantive standard, audience promise, format, or publication-readiness criterion | The governing `STYLE.md` |\n| Confirmed reader situation, knowledge, interest, want, resistance, or audience research finding | The resolved `AUDIENCE.md`, including a named shared source; retain source, date, and scope |\n| Repeated editorial failure | Relevant guardrail or review reference |\n| Reusable workflow improvement | Workflow or architecture file |\n| Piece-specific decision or open loop | The active draft, notes, or project status artifact |\n\nNever save durable learning into an installed plugin/cache copy. Do not append lessons to a skill's `SKILL.md` merely because the skill was active.\n\nIf a legacy `TASTE.md` is still present, route new learning to `VOICE.md`, `STYLE.md`, or the governing `AUDIENCE.md` and offer migration. Do not delete or archive `TASTE.md` without explicit approval.\n\nWhen one correction mixes sentence craft and substantive standards, split it into two atomic rules. Do not duplicate the combined rule across both files.\n\n## Decide Whether It Should Generalize\n\nDistinguish:\n\n- an explicit enduring preference;\n- a repeated pattern supported by several examples;\n- a one-off choice for this piece;\n- an inference that still needs confirmation.\n\nIf the user explicitly says to save an enduring preference, proceed once the destination is clear and permitted. If the lesson is inferred, show the proposed rule and ask the user to confirm before making it durable.\n\nFor audience learning, a piece-specific assumption stays with the assignment. Do not promote positioning or an editorial inference to observed reader behavior. A shared audience guide may affect multiple publications; preserve its scope and update it only within the user's authorization. Follow the existing shared source rather than creating a competing local copy.\n\n## Write Precisely\n\n1. Read the destination file before editing.\n2. Add or revise the narrowest relevant rule.\n3. Preserve existing user-authored context, examples, and avoid duplicate guidance.\n4. Record the mechanism or consequence, not only a vague preference.\n5. Keep examples fresh and non-copyable when the rule concerns voice.\n\n## Confirm\n\nReport the exact rule captured and the file or system updated. If a higher-authority destination requires approval, provide the proposed wording and wait instead of pretending it was saved.\n"
}SHA-256 of public snapshot: 752610314250a13c24834c8cb18ee8c881429738bcbe6fa1ff10d62f31b020c5