mirror of
https://github.com/larksuite/cli.git
synced 2026-07-03 14:02:43 +08:00
c15cb120a1
Regenerate lark-im SKILL.md (4253 -> 2040 tokens, tiktoken cl100k_base, -52%) via per-domain gen-skills flags. Drop the scope permission table and the duplicated schema-usage note, compact API Resources to a terse identity index, render the Shortcuts table as one-line routing, and move Shortcuts ahead of the concept sections (executable-first). Full detail stays in --help / schema (the source of truth); Go Desc and --help are unchanged. Add the NOT boundary, a good/bad sender example, Flag-type rationale, and anti-hallucination guidance with a safe fallback (guide users to the Feishu client; do not auto-execute high-risk raw-API writes). The master template is variabilized with defaults preserving current output, so other domains regenerate byte-identical (verified). Eval (skillave, IM, 9 cases x 3 runs, lark-cli-eval-analyze two-table compare): mean pass_rate 0.939 -> 0.969, duration -22%, tool-calls -26%, zero structural regression; anti-hallucination case 0.58 -> 1.00. Depends on larksuite-cli-registry change (skill-meta.yaml per-domain flags + gen-skills.py); must merge together with (or before) this PR. Change-Id: I33c62664388542bfd8ba85ea06c1c5a493c0d935