# issue 17110 Claude effort-cap cross-check (sanitized: applied effort + T3 label only) # Claude Code CLI 2.1.303 (Claude Code) # @anthropic-ai/claude-agent-sdk 0.3.276 # model claude-opus-5-5, requested effort max, isolated HOME/CLAUDE_CONFIG_DIR, PATH only # T3 commit: 3e836a5a1b7b4a927beeb0a2592e2f24161f0c98 (fix on top of 98f75a67a7) MATCH no caps cli=max t3=max maxLabel="Max" MATCH top-level high cli=high t3=high maxLabel="Max (capped to High)" MATCH alias opus low cli=low t3=low maxLabel="Max (capped to Low)" MATCH alias OPUS[1M] medium cli=medium t3=medium maxLabel="Max (capped to Medium)" MATCH dated canonical medium cli=medium t3=medium maxLabel="Max (capped to Medium)" MATCH dotted claude-opus-5.5 ignored cli=high t3=high maxLabel="Max (capped to High)" MATCH dotted opus-5.5 ignored cli=max t3=max maxLabel="Max" MATCH bedrock id low cli=low t3=low maxLabel="Max (capped to Low)" MATCH less specific claude-opus-5 cli=max t3=max maxLabel="Max" MATCH per-model max exempts own top cli=max t3=max maxLabel="Max" MATCH max key doesn't beat low key cli=low t3=low maxLabel="Max (capped to Low)" MATCH user max vs project top high cli=high t3=high maxLabel="Max (capped to High)" MATCH user top medium vs project opus high cli=medium t3=medium maxLabel="Max (capped to Medium)" MATCH flag settings dated low cli=low t3=low maxLabel="Max (capped to Low)" MATCH other family sonnet ignored cli=max t3=max maxLabel="Max" # Result: 15/15 MATCH # Linear-time benchmark of the key resolver on malformed input: # 1 key, near-miss prefix x1000 keys= 1 chars= 14001 0.15ms # 1 key, near-miss prefix x10000 keys= 1 chars= 140001 0.76ms # 1 key, near-miss prefix x100000 keys= 1 chars= 1400001 7.02ms # 3000 malformed keys keys= 3000 chars= 88884 4.36ms # 30000 malformed keys keys=30000 chars= 918885 25.81ms # 3000 tokens one key ("opus [1m] " repeated) keys= 1 chars= 30000 0.01ms # 30000 tokens one key ("opus [1m] " repeated) keys= 1 chars= 300000 0.29ms # 1 key, 1M chars of '[' keys= 1 chars= 1000004 0.99ms