tune-outputlisted

Change how Claude's replies are shaped — adopt, revise or reject a response-style rule — then record the pass in a ledger so the next one starts from evidence instead of from scratch. Owns the output style file, the blind A/B harness that tests a candidate before it ships, and the history of what has already been tried. TRIGGER when: the user wants replies formatted or worded differently, brings a response-style idea from another project or repo to evaluate, asks why a style rule exists or is missing, or says an existing style rule is not working. DO NOT TRIGGER when: the subject is prose inside a document or README (that is `documentation`), commit message wording (`shared/commit-message-rules.md`), a single one-off reply the user just wants rewritten, or a subagent's prompt.
AnotherSava/claude-code-common · ★ 1 · AI & Automation · score 74

Install: claude install-skill AnotherSava/claude-code-common

# Tune output ## Context - Current state: !`bash ~/.claude/skills/tune-output/scripts/preflight.sh` ## What this skill is for Response shape is tuned repeatedly and each pass forgets the last one. The cost is not the wasted work — it is re-adopting a rule that was already measured harmful, or re-arguing a rejection whose reason nobody wrote down. The ledger is the point of this skill; the harness exists to give the ledger something true to record. ## Process ### 1. Read the ledger before forming any opinion Read `~/.claude/skills/tune-output/ledger.md` in full. It is short by design. If the candidate rule — or its near-twin — already has an entry, **say so before proposing anything** and open with what the last pass concluded. A rule previously rejected is re-opened only with new evidence, and the entry names what evidence would qualify. Never silently re-adopt something the ledger rejected. ### 2. Locate the rule's home An output rule can live in four places, and picking the wrong one is the most common failure. Use **Current state** from Context to see what is configured now, then choose: | The rule is… | Home | |---|---| | how a reply is shaped, always | the output style — `claude/output-styles/<name>.md` | | a word or phrase never to use | the **Overused Phrases** section of `claude/CLAUDE.md` (it already has the entry-plus-replacement format) | | a preference with a *why* worth keeping | a `claude/memory/feedback_*.md` file, indexed in CLAUDE.md | | something