resolve-issue-learningslisted

# Resolve issue learnings Turn the generic, cross-repo observations that `resolve-issue` quietly captured while running — about its own gate flow, step sequencing, state handling, and how it chains its component skills — into learnings that actually improve future runs. This skill is the *harvest* half of a capture-then-harvest loop: `resolve-issue` captures cheap, unverified candidates during every run (see its own "Capturing generic learnings" section); this skill verifies them against the current skill as ground truth and applies only the ones that survive. The split is on purpose — the capture moment (mid-run, sample-of-one, no skill source on hand) is the worst time to decide a learning is true, so nothing is trusted until it is verified here. It runs **two ways** — automatically, as `resolve-issue`'s own internal upkeep (a safe machine subset, no human in the loop), and deliberately, when a person invokes it by hand (the full version, including the human-judged backlog and promotion to the shipped skill); see Invocation below. It has **two modes**, and the mode is chosen by *what is reachable*, never by a folder or repo name (those are unreliable — a checkout's folder and its remote repo name routinely differ): - **Mode X (default, any user, any repo)** — write the verified learnings to a **user-global conventions file** that `resolve-issue` reads at the start of its next run and honors when consistent. Fully local, self-contained, no outward action. This is what mak