Read a Tool Row Without a Trailing Pipe, and Report the License Column

[ptr727]

Raised by Copilot on the promotion pull request #635, against develop code rather than against anything that promotion introduces. Both halves of the finding are real, and both are measured below rather than argued.

The trailing pipe

_TOOL_ROW required a closing | after the description. GitHub's Markdown makes that pipe optional, so a valid row written without one matched nothing and was skipped by continue before any check ran on it:

MATCH '| [cspell][cspell-link] | Spell checker. |' -> 'Spell checker.'
SKIPPED '| [cspell][cspell-link] | Spell checker.' -> None
MATCH '|[cspell][cspell-link]|Spell checker.|' -> 'Spell checker.'
MATCH '| [cspell][cspell-link] | Spell checker. | MIT |' -> 'Spell checker.'

A skipped row loses its link check, its description check, and its place in the ordering list, so a whole table written that way reports zero findings. That is the failure shape this dimension exists to avoid: it scores as clean rather than as unread.

Nothing else catches it either. MD060 is disabled outright in .markdownlint-cli2.jsonc, and MD055 defaults to consistent, which passes a table whose rows all omit the closing pipe. So the form is permitted everywhere and was read nowhere.

The License column

spec/project-types.json already asserted "The section carries no license column" at verdict letter, and nothing read it. spec/readme-structure.md states the rule and the reason: a license belongs to the dependency and is authoritative at its source. This makes the existing assert honest rather than adding a new rule.

Two decisions bound it:

• Read off the header, not the rows, so it reports once for the table rather than once per tool, and so a tool whose own role is the word "License" is not mistaken for the column. There is a test in that direction.
• That one column and no other. Copilot suggested flagging extra columns generally; the spec forbids the License column and says nothing about any other, so a general check would grade the fleet on a rule nobody wrote.

A row carrying the extra column still parses correctly, because the description capture is non-greedy, so the per-tool checks keep working alongside the new finding.

Measured against the corpus

All 22 cataloged READMEs, at their ground-truth branches, old check against new, read in both directions:

	Old	New
Blog	4	5
LanguageTags	12	12
MediaTools	13	13
PlexCleaner	27	27
Total	56	57

One finding added, Blog's License column, and no verdict lost in the other direction. The categories break down as 48 absent descriptions, 4 links, 3 descriptions, 1 ordering, 1 License column.

The honest half of that result: no repo in the fleet writes a trailing-pipe-less row today, so that fix closes a blind spot rather than a live miss. It is worth taking anyway, because the blind spot is silent and the 17 repos that still owe this section have not written their tables yet.

TODO.md takes the new count, and two pre-existing miscounts in the same bullet are corrected while I am in it: PlexCleaner's 27 is its total rather than its absent-description count, which is 25, and all three of Blog's roles differ from the catalog rather than two, since Hugo's matches in substance but differs by its opening capital and its full stop.

Verification

spec/audit.py --selftest passes with 7 new cases, covering both halves in both directions: a row read without a trailing pipe, one judged without one, a whole table ordered without them, the License column reported once, the rows still read alongside it, a table without the column not reported, and a row cell reading "License" not mistaken for the header. spec/validate.py, prose_lint.py --diff develop, repo_gate.py --check eol-coverage, editorconfig-checker, markdownlint-cli2 and cspell all clean. spec/audit.py stays LF, TODO.md stays CRLF.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR tightens the README audit’s parsing of the 3rd Party Tools section so valid Markdown table rows aren’t skipped, and adds a new rule to report a forbidden License column based on the table header. It also updates TODO.md to reflect the newly measured finding counts and corrected breakdown.

Changes:

• Update _TOOL_ROW parsing to accept tool-table rows that omit the trailing |.
• Add header-based detection of a License column in the 3rd Party Tools table, plus selftests for the new behaviors.
• Refresh TODO.md measurement dates/counts and correct prior miscounts.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
TODO.md	Updates the recorded fleet measurement totals/dates and corrects the described breakdown for 3rd Party Tools findings.
spec/audit.py	Improves Markdown table parsing for 3rd Party Tools rows and adds a header-based License-column finding with expanded selftests.

Suppressed comments (1)

spec/audit.py:710

• table_cells() only recognizes rows that start with |. GitHub-flavored Markdown also allows header/separator/data rows without the leading pipe (e.g. Tool | Role). With the new License-column check keyed off the header, a table using that valid form won’t be recognized and can again score as clean despite being present. Expanding table_cells() to accept rows with or without the leading pipe avoids that blind spot.

 s = line.strip()
if not s.startswith("|"):
return None
s = s[1:-1] if s.endswith("|") else s[1:]
return [c.strip() for c in s.split("|")]

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[ptr727]

Answering the suppressed comment on spec/audit.py:710, which carries no thread to resolve:

table_cells() only recognizes rows that start with |. GitHub-flavored Markdown also allows header/separator/data rows without the leading pipe (e.g. Tool | Role). With the new License-column check keyed off the header, a table using that valid form won't be recognized and can again score as clean despite being present.

Accepted, and fixed in 45fde3e together with the threaded finding it duplicates. It is the sharper statement of the two, because it names the consequence exactly: the License column is read off the header, so a header written without a leading pipe would have taken the check straight back to scoring clean on a table nothing had read.

table_cells now requires a separating pipe rather than a leading one, and drops each outer pipe only where it is present. The separator is doing real work: --- alone carries no pipe and returns None, so a thematic break is not a one-cell delimiter row and the paragraph above it is not read as a table header. Selftest cases cover that direction and prose naming a cataloged tool.

Re-measured across all 22 cataloged READMEs, in both directions: the total stays 57 and no verdict moves, so no false positive. Worth being plain about the other half of that result, though: neither outer-pipe form appears anywhere in the fleet today, so both fixes close a silent blind spot rather than catch a live miss. The 17 repos that still owe this section have not written their tables yet, which is the reason to take it now rather than after.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.