Self-check item 8 in scriptorium/skills/docs/SKILL.md ("All code fences carry a language and are complete") is the one item in the nine-item emission gate with no mechanical coverage at all. The skill says so plainly today: the linter "covers part of item 7 and part of item 9, and none of item 8: it never checks whether an info string is present."
Deliberately left out of PR #42 (closing #27), which was a spec change; its ## Out of scope names this as follow-up.
The change is small; the ripple is why it was deferred
scriptorium/skills/docs/scripts/check_markdown.py already captures what it needs. scan_fences stores Fence.info_string, and the only consumer is check_backtick_info_string, which fires on a backtick within an info string and never on its absence. The check is roughly if not fence.info_string.
What makes it more than three lines:
LLMS_DESCRIPTION states the linter's coverage by count ("five error-severity checks", "five advisory-severity checks"). A sixth changes both the text and the shape it advertises.- The 52-test suite in
scripts/test_check_markdown.py asserts against that shape.
scriptorium/skills/docs/references/commonmark.md carries a ## Scope note documenting the same coverage, and SKILL.md documents the item-8 gap as permanent. Both need updating, or they become the wrong half of a contradiction.
Decide before implementing
Error or advisory? Error severity makes exit 1 stop emission, which is what item 8 being a gate item implies. Advisory keeps the exit code clean but leaves item 8 effectively unenforced. The repo's own markdown must pass at whichever severity is chosen.
Worth checking against a related finding: the daedalus altitude review noted that self-check items 2 through 5 are all decidable by a program over the file path and the YAML block, and that eval assertions already state some of them as regexes. If a check_frontmatter.py is coming, this check may want to land alongside it rather than alone.
Acceptance Criteria
Self-check item 8 in
scriptorium/skills/docs/SKILL.md("All code fences carry a language and are complete") is the one item in the nine-item emission gate with no mechanical coverage at all. The skill says so plainly today: the linter "covers part of item 7 and part of item 9, and none of item 8: it never checks whether an info string is present."Deliberately left out of PR #42 (closing #27), which was a spec change; its
## Out of scopenames this as follow-up.The change is small; the ripple is why it was deferred
scriptorium/skills/docs/scripts/check_markdown.pyalready captures what it needs.scan_fencesstoresFence.info_string, and the only consumer ischeck_backtick_info_string, which fires on a backtick within an info string and never on its absence. The check is roughlyif not fence.info_string.What makes it more than three lines:
LLMS_DESCRIPTIONstates the linter's coverage by count ("five error-severity checks", "five advisory-severity checks"). A sixth changes both the text and the shape it advertises.scripts/test_check_markdown.pyasserts against that shape.scriptorium/skills/docs/references/commonmark.mdcarries a## Scope notedocumenting the same coverage, andSKILL.mddocuments the item-8 gap as permanent. Both need updating, or they become the wrong half of a contradiction.Decide before implementing
Error or advisory? Error severity makes exit 1 stop emission, which is what item 8 being a gate item implies. Advisory keeps the exit code clean but leaves item 8 effectively unenforced. The repo's own markdown must pass at whichever severity is chosen.
Worth checking against a related finding: the daedalus altitude review noted that self-check items 2 through 5 are all decidable by a program over the file path and the YAML block, and that eval assertions already state some of them as regexes. If a
check_frontmatter.pyis coming, this check may want to land alongside it rather than alone.Acceptance Criteria
check_markdown.pyflags a code fence with no info string, at a severity chosen and stated in the issue threadLLMS_DESCRIPTIONreports the new check and its corrected countstest_check_markdown.pycovers the new check both ways, and its shape assertions match the new countsreferences/commonmark.md## Scope noteandSKILL.md's item 7-9 paragraph both describe the new reach; the "none of item 8" claim is removedbash scripts/gates/structure.shandclaude plugin validate .both exit 0