You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Free checks: rules size limit and eval tag hygiene #3
Two more standard-library checks for checks/, run by make validate:
The combined rules text that hooks/load-rules.sh prints must stay under the 10,000-character hook output limit; Claude Code truncates silently past it. It is 4,977 characters at the time of writing.
Every eval case under evals/ must carry its component's name as a tag, because the make eval-* targets select cases with --tag and an untagged case is silently never run. All 20 cases are tagged today; the check keeps it that way.
Planning note (2026-10-13): done in one combined PR with #1, #3, #6, #7 and #15, after #11 lands. #3's size check goes in first within that PR so the rule additions from the others are measured against the 10,000-character limit. #7 waits for #11 because it quotes the pr skill's wording. #15 is taken as the orientation sentence in the code-changes rule; the maintainer's-kit plugin stays as the fallback on that issue. One spec task per issue, one Changes bullet per issue.
Status (2026-10-13): the rules-size half of this issue shipped in #21 as checks/test_rules_size.py, which measures the hook's output against the 10,000-character limit and rules/collaboration.md against its own limit, and runs under make validate.
The eval-tag half ships with the 3-eval-checks-and-cadence branch as checks/test_eval_cases.py, which also checks every grader's type and required fields and every scaffold's wiring and executable bit. That PR closes this issue.
Activity
sabind commentedon Oct 13, 2026
Planning note (2026-10-13): done in one combined PR with #1, #3, #6, #7 and #15, after #11 lands. #3's size check goes in first within that PR so the rule additions from the others are measured against the 10,000-character limit. #7 waits for #11 because it quotes the pr skill's wording. #15 is taken as the orientation sentence in the code-changes rule; the maintainer's-kit plugin stays as the fallback on that issue. One spec task per issue, one Changes bullet per issue.
sabind commentedon Oct 13, 2026
Status (2026-10-13): the rules-size half of this issue shipped in #21 as
checks/test_rules_size.py, which measures the hook's output against the 10,000-character limit andrules/collaboration.mdagainst its own limit, and runs undermake validate.The eval-tag half ships with the
3-eval-checks-and-cadencebranch aschecks/test_eval_cases.py, which also checks every grader's type and required fields and every scaffold's wiring and executable bit. That PR closes this issue.Merge pull request #22 from freethinkingit/3-eval-checks-and-cadence