WolfMarkDown v1.0.3

WolfMarkDown v1.0.3 makes Markdown cleanup more reliable. It strengthens the checks that help catch lost structure, improves citation repairs, and makes verification results easier to review.

What’s improved

• Better protection for meaningful numbering. Stronger checks catch rankings and continuation lists that have been accidentally renumbered during cleanup.
• More reliable source lists. Checks help catch records that have collapsed into a paragraph, changed order, or become separated from their reference labels.
• Safer citation cleanup. Optional repairs separate citations that run together while protecting code examples and valid Markdown links.
• Clearer verification reports. Results distinguish what passed, failed or was skipped, show the formatting rules in use, and flag unusually large reductions in document lines for review.
• Less disruptive maintenance. Routine dependency repairs leave the agent's existing skill setup in place.

WolfMarkDown still keeps document structure and meaning under the agent's judgement. Its checks provide evidence about the Markdown properties they cover; a PASS is not a fact-check or approval of the content.

Get the update

Update an existing Skills CLI installation with:

npx skills update wolfmarkdown

Reload your agent host after updating. For first-time installation, repository checkouts and other update options, see Installation and updates.

Node.js 22 or newer is required.

Validation

This release passes 279 automated tests on Node.js 22 and 24, Agent Skill validation, and verification of the public Markdown example.

All changes since v1.0.2.

WolfMarkDown v1.0.2

Overview

WolfMarkDown v1.0.2 focuses on correctness, safety, reliability, and stronger verification. Integrity comparison now understands what it protects, publishing is safe when multiple agents operate concurrently, and every operation leaves structured evidence behind.

Highlights

• Stronger preservation of technical content: code spans, URLs, currency, paths, environment assignments, and image descriptions fail verification when changed or removed, while ordinary prose edits raise review warnings instead of failures.
• Improved detection of altered values and identifiers, including same-class value swaps, occurrence shifts, and accurate occurrence counts in receipts.
• More reliable publishing when multiple agents operate concurrently: per-destination locks, stale-lock recovery, conflict refusal instead of silent overwrites, and temporary candidates published only after verification passes.
• Accurate verification receipts and clearer error reporting: truthful hashes, stage-specific read errors, per-file error receipts, skipped-path reporting, and failure evidence for batches with no eligible files.
• Improved skill installation and diagnostics: Node.js pre-checks, classified npm failure messages, link-only setup, and a Doctor exit code that follows runtime health with --strict audits.

Compatibility

• Node.js 22 or newer is required.
• Existing installations should update dependencies (npm ci --prefix wolfmarkdown) and re-run setup.
• Doctor's exit code is now 0 when the runtime is healthy, even if discovery links are missing; use node wolfmarkdown/scripts/doctor.mjs --strict for a full installation audit.
• node wolfmarkdown/scripts/install.mjs --skip-deps manages discovery links without running npm ci.

Installation and updating

Agent Skills CLI (recommended):

npx skills add WolfMarkTools/WolfMarkDown --skill wolfmarkdown

Update an existing Skills CLI installation:

npx skills update wolfmarkdown

Repository checkout:

git clone https://github.com/WolfMarkTools/WolfMarkDown.git
cd WolfMarkDown
npm ci --prefix wolfmarkdown
node wolfmarkdown/scripts/install.mjs

Update a checkout with git pull, then repeat npm ci --prefix wolfmarkdown and node wolfmarkdown/scripts/install.mjs.

Verification

• 204 automated tests pass on Node.js 22 and 24.
• Agent-skill metadata validation passes.
• The public Mobile App Migration example verifies PASS against its source snapshot.

Known limitations

• Integrity checks cover recognised token classes only: some values (two-part versions, month-name dates, isolated integers) are outside the extractor and reported as coverage warnings.
• A PASS is Markdown-quality evidence — well-formed, verified output — not factual correctness, completeness, policy compliance, or authorisation to publish.
• npm audit reports one low-severity KaTeX advisory chain via markdownlint; WolfMarkDown never renders math to HTML, so the documented rendering conditions do not apply. Tracked upstream rather than forcing a breaking downgrade.

Changelog

• Full comparison: v1.0.0...v1.0.2
• Pull request: #13

WolfMarkDown v1.0.0

Introducing WolfMarkDown v1.0.0

WolfMarkDown v1.0.0 is the publishing workflow for agent Markdown, not a patch on v0.2.2.

The agent still decides structure. Deterministic tooling still proves the file is safe to keep. What v1.0.0 adds is written evidence: a receipt, a source-to-candidate preview, integrity coverage, and publish paths that restore or refuse instead of leaving a half-written .md.

A PASS is Markdown-quality evidence. It is not factual correctness, completeness, or approval to publish.

Highlights

• Receipts. Verify records hashes, tool versions, checks, structured issues, remediations, and the quality boundary. A PASS is durable evidence, not only an exit code.
• Preview before keep. Source-to-candidate preview inventories observable structure, recognised protected tokens, and whether the candidate is formatting-only or content-changed.
• Honest integrity. Coverage warnings name shapes the extractor does not protect, including two-part versions such as 18.17, month-name dates such as May 2027, and isolated integers.
• Batch Verify. Directories, multiple files, and standard input are supported.
• Long-document scaffold. The agent gets an inventory of headings, fences, lists, tables, and recognised tokens before it decides structure. Scaffold does not rewrite the document.
• Safer writes. Format and publish use atomic replace, refuse symlink destinations, and will not overwrite a file that changed during verification. Failed Clean restores the original. Failed Compose does not publish.

Compose, Clean, Verify, Doctor, and Setup remain the public operations.

Install and update

Requires Node.js 20 or newer.

npx skills add WolfMarkTools/WolfMarkDown --skill wolfmarkdown

npx skills update wolfmarkdown

Reload the host after install or update. A repository checkout still updates with git pull, npm ci --prefix wolfmarkdown, and node wolfmarkdown/scripts/install.mjs.

Behaviour to know

• Clean and Compose now keep preview and receipt evidence as part of keep-or-publish. Bare Verify remains available for check-only use.
• The skill activates for Markdown, .md, export/save, or WolfMarkDown itself. Generic “clean up this research” or “format this documentation” does not start Clean or Compose.
• Standalone Verify without --integrity-from does not prove protected tokens were preserved.
• CI runs the test suite, skill validation, and the public example on Node 20, 22, and 24.

What's Changed

Compared with v0.2.2, v1.0.0 is better where it used to stop at “the command exited zero”:

• v0.2.2 proved a file and moved on. v1.0.0 writes a receipt you can keep: hashes, versions, checks, issues, remediations, and the quality boundary.
• v0.2.2 integrity could pass while leaving unprotected shapes silent. v1.0.0 warns when two-part versions, month-name dates, or isolated integers were never in the extractor.
• v0.2.2 Verify was one file at a time. v1.0.0 verifies directories, multiple files, and standard input, with a source-to-candidate preview before you keep the result.
• v0.2.2 left long-document structure to the agent’s memory. v1.0.0 adds a scaffold inventory so headings, fences, lists, and tables are mapped before repair.
• v0.2.2 publish was less strict about collisions and in-flight edits. v1.0.0 uses atomic replace, refuses symlink destinations, will not clobber a file that changed during verification, restores on failed Clean, and does not publish failed Compose.
• v0.2.2 README assumed you already knew how to install. v1.0.0 makes install and update obvious for both the skills CLI and a checkout.
• v0.2.2 could activate on generic research or docs cleanup. v1.0.0 starts Clean or Compose only when the request is Markdown, a .md file, export/save, or WolfMarkDown itself.

Ship WolfMarkDown v1.0.0: #11

Full Changelog: v0.2.2...v1.0.0

WolfMarkDown v0.2.1

What's Changed

• v0.2.1: polish agent publishing workflow by @WolfMarkTools in #7
• docs: restore README product context after v0.2.1 by @WolfMarkTools in #8
• Add mobile migration example fixtures by @WolfMarkTools in #9

Full Changelog: v0.2.0...v0.2.1

WolfMarkDown v0.2.0

WolfMarkDown v0.2.0

WolfMarkDown now goes beyond valid Markdown formatting to recover clear document structure from messy pasted material.

Highlights

• Recovers latent headings, sibling sections, labelled lists, ordered steps, paragraphs, and comparison tables when source evidence is clear.
• Rebuilds GFM tables only when both column boundaries and a recognisable header row are present.
• Preserves ambiguity rather than inventing a table, heading, list, schema, or claim.
• Adds long-document handling with a document ledger and semantic handoff for cross-section reconciliation.
• Preserves protected technical identifiers and prevents unsafe publication of failed or unverified output.
• Adds synthetic regression coverage for structural recovery and conservative ambiguity preservation.
• Clarifies that a WolfMarkDown PASS is Markdown-quality evidence, not factual validation, policy approval, or publication authorisation.

Compatibility

WolfMarkDown is model-agnostic and works with Agent Skills hosts that load .agents/skills/wolfmarkdown or .claude/skills/wolfmarkdown.

Node.js 20+ is supported, including Node.js 22 and 24.

Verification

• 79/79 Node tests passing
• skills-ref validate ./wolfmarkdown passing
• Prettier, markdownlint, GFM parsing, protected-token integrity, and idempotence checks passing

Notes

Semantic repair remains agent-judged. WolfMarkDown provides the workflow, preservation rules, and deterministic proof; the host agent remains responsible for applying semantic judgement faithfully.

What's Changed

• docs: turn README into product landing page by @WolfMarkTools in #4
• fix: improve semantic structure recovery for pasted reports by @WolfMarkTools in #5

Full Changelog: v0.1.1...v0.2.0

WolfMarkDown v0.1.1

Add Agent Plugins 1.0 packaging for marketplace distribution. No WolfMarkDown behaviour changes.

What's Changed

• chore: add Agent Plugins packaging by @WolfMarkTools in #3

Full Changelog: v0.1.0...v0.1.1

WolfMarkDown v0.1.0

WolfMarkDown v0.1.0

Initial public release of WolfMarkDown, a cross-agent Markdown output and repair skill.

Highlights:

• Create/export verified Markdown files
• Clean and restructure existing Markdown
• Remove copied AI conversation scaffolding
• Repair malformed headings, lists and tables
• Preserve technical identifiers, URLs and citations
• Deterministic Prettier formatting
• markdownlint and GFM validation
• Integrity and idempotence checks
• Failure-safe Clean and Compose workflows
• Setup and Doctor support
• Agent Skills compatibility across major coding-agent harnesses

Install:

npx skills add WolfMarkTools/WolfMarkDown --skill wolfmarkdown