tune

Solid

Process the issue-to-pr friction log into concrete skill/script improvements. Reads .claude/issue-to-pr/friction.log, proposes batched edits with evidence, and applies them on your approval. Run it when the log has built up.

AI & Automation 6 stars 0 forks Updated 4 days ago MIT

Install

View on GitHub

Quality Score: 81/100

Stars 20%
28
Recency 20%
100
Frontmatter 20%
70
Documentation 15%
88
Issue Health 10%
80
License 10%
100
Description 5%
100

Skill Content

# tune — turn friction into fixes The issue-to-pr pipeline no longer has an always-on "improve this skill" step. Instead each run appends one line to `.claude/issue-to-pr/friction.log` (repo-local, gitignored) whenever a step genuinely fought back — a stop that should not have happened, an unclear instruction, a missed detection, a script that surprised the model. This skill batches that evidence into real changes. ## Steps 1. **Read the log.** `.claude/issue-to-pr/friction.log` (from the repo root). Each line is a dated, one-sentence friction note with the step it hit. If the file is absent or empty, say so and stop — nothing to tune. 2. **Cluster.** Group the lines by root cause, not by surface symptom. Three notes about the same confusing exit code are one fix, not three. 3. **Propose.** For each cluster, propose the smallest concrete change to the SKILL, a script, or a reference — quoting the friction lines as evidence. A change with no evidence line does not belong here. Prefer edits that make a stop mechanical (an exit code / a test) over prose that asks the model to remember. 4. **Confirm, then apply.** Show the batched proposal and ask for a go-ahead. On approval, make the edits (bump the plugin version + changelog per repo rules), run the gates (`tests/run-tests.sh`), and — if the changed area is behavioral — note that a fresh dogfood run should confirm it. With `--dry-run`, stop after the proposal. 5. **Prune.** After applying, remov...

Details

Author
DmitriyYukhanov
Repository
DmitriyYukhanov/claude-plugins
Created
6 months ago
Last Updated
4 days ago
Language
Shell
License
MIT

Integrates with

Bundled in these plugins

Similar Skills

Semantically similar based on skill content — not just same category

AI & Automation Listed

skill-tuning-loop

Empirically test whether a lacquer skill needs an edit by mining real session transcripts for friction, proposing a bounded fix, and validating it against held-out cases before it ships. Use when a skill in core/skills/ or profiles/*/skills/ keeps getting corrected in practice, when triaging a scheduled tuning-pass PR, or when skill-authoring-standard's manual rubric isn't enough to tell whether a proposed edit actually helps. Distinct from skill-authoring-standard (a static prose rubric checked by a human) — this is an empirical, evidence-gated loop checked by rollouts.

1 Updated today
patrickserrano
AI & Automation Listed

tune

Harness retro — turn recurring agent mistakes into one-line fixes. Mines recent session transcripts (plus the bugfix log and past runs) for repeated corrections, permission friction, and rules that get violated anyway, then proposes ranked changes one at a time: a CLAUDE.md rule, an allowlist entry, a script replacing repeated prose, a new-skill candidate, or a deletion that no longer earns its context cost. Nothing applies without approval; accepted/rejected findings persist in out/tune/learnings.md. Use when the user says "/tune", "improve my setup", "why does claude keep doing X", or after a frustrating session. done's one-line retro catches one lesson at ship time; tune is the periodic audit of the whole harness.

6 Updated today
duthaho
AI & Automation Listed

claudemd-review

Use right after being corrected on something a CLAUDE.md file should already have told you, when explicitly asked to review CLAUDE.md, or periodically after a session with a lot of back-and-forth — reviews THIS conversation for friction and proposes a bounded CLAUDE.md edit. A lightweight, single-session complement to skill-tuning-loop (which mines many sessions and validates a proposal against held-out cases before surfacing it — reach for that instead when the ask is "make this systematic," not "check right now"). Own version of the dx plugin's `/dx:review-claudemd`.

1 Updated today
patrickserrano