| name | ma-wiki-lint |
| description | Periodic tidy pass on the wiki/MEMORY tree at `.madagents/wiki/` — structural checks, merge near-duplicates, split overgrown pages, balance the MEMORY/wiki tiers, prune; plus a frontmatter scan sub-operation. Reorganizes already-recorded content; it does not source-walk, run MadGraph, or author new principles. User-invoked; do not auto-invoke at session start or end. |
/ma-wiki-lint [targets]
A periodic tidy of the learned tier — the wiki at .madagents/wiki/ and each owner's MEMORY.md. It keeps what is already recorded well-formed, deduplicated, and balanced across the two tiers (MEMORY.md slate and wiki bodies). It reorganizes existing, already-cited content — it never source-walks, runs MadGraph, or authors a new principle.
Learning new content is a different job and lives elsewhere: ma-wiki-write records a finding on insight; generalization (lifting a deeper principle that catches more cases) fires on insight during mg-study / ma-reflect / mid-work via ma-wiki-write's generalize, never here; correctness of cached content is held by grounding-at-write, ma-reflect on a surfaced mistake, and /mg-deep-verify at point-of-use.
User-invoked only. Do not auto-invoke at session start or end.
Target
If the user did not name targets, ask: lead, named consultants, or all. Skip if .madagents/wiki/ is empty. Each owner tidies its own subtree — the memory-carrying agents (consultants and the probe) dispatched in parallel, the lead concurrent on lead/; no owner reads another's subtree.
Per-owner dispatch (tidy-mode)
"Tidy-mode dispatch on .madagents/wiki/consultants/<your-agent-name>/ (literal subtree path, suffix and all). Reorganize what is already recorded — do not source-walk, do not run MadGraph, do not author a new principle. Cheap fixes you apply via ma-wiki-write; findings beyond your reach you list.
Structural:
(a) every page has parseable description: frontmatter and only that field — flag/fix other keys;
(b) descriptions ≥40 chars and not a trivial restatement of the filename;
(c) no two pages share the same description (semantic duplicate → merge, below);
(d) filenames not in blocklist (notes.md, misc.md, general.md, temp.md, todo.md, wip.md, dated filenames);
(e) directory name exactly matches your agent's filename, suffix and all (e.g. ma-…-consultant, or ma-probe);
(f) MEMORY.md exists and is within budget (≤80 lines target, ≤120 hard cap; if over, demote stale entries);
(g) MEMORY.md's wiki page index matches the actual pages in your subtree (fix missing/stale entries).
Consolidate (reorganize existing content only — no new claims):
(h) Merge two pages with the same trigger + behaviour, differing only in examples → one page;
(i) Split a page whose body cleanly separates into two scopes that shouldn't co-load on one query;
(j) Tier-balance — promote a wiki page's distillation into MEMORY.md when its principle is operative on most recent dispatches; demote a MEMORY.md entry that is internalized or long-unused.
Cross-subtree refs are not yours to update — flag them.
Return: clean / N findings; one line per finding with the ma-wiki-* operation applied; cross-subtree refs flagged."
Lead pass
The lead runs its own tidy on lead/ directly, adding: flat tree (no subdirs); consultant directory list matches the real consultant agents.
Routing pass. Refine the description of each dispatched consultant whose boundary sharpened, and the index entry of each lead/ page you touched. Then sweep the whole roster — not only what you refined: every description valid frontmatter, under the 800-character cap, and free of the two description bugs.
Refining a description needs input the lead does not hold, so ask for it: a consultant's tidy-mode return should say which regimes and situations properly belong to it, and which it should hand off — drawn from its own wiki and slate.
Aggregate
Per owner: (1) clean / N findings; (2) one line per finding with the recommended ma-wiki-* operation; (3) cross-subtree refs flagged.
Scan sub-operation
A frontmatter description scan over a wiki subtree: returns one line per page — <path> description: <text>. It is read-only — it returns descriptions and reorganizes nothing (tidying is owner-only); a reviewer or auditor uses it purely to orient on a subtree it does not own. This is the fallback for when MEMORY.md is not auto-loaded — for normal subagent dispatches the auto-loaded MEMORY.md already carries the index of pages in the agent's own subtree.
When you actually need it
For memory-carrying agents (the consultants and the probe) and the lead, MEMORY.md auto-loads at every dispatch and already carries the index. Use the scan only when:
- You're a reviewer /
ma-blind-spot-auditor / ma-failure-mode-extractor scanning another agent's subtree for orientation (those agents have no auto-memory of the consultant they verify).
- You need to refresh the index manually (rare; auto-memory does this automatically).
- The auto-memory loaded fewer entries than the wiki contains (overflow case; auto-memory caps at first-200-lines).
Procedure
find <subtree> -maxdepth 2 -name '*.md' \
-exec awk '/^description:/{print FILENAME ": " $0; nextfile}' {} +
Report verbatim. Empty subtree → empty.
Read-scope discipline
- Lead scans
lead/ only. To get a consultant's content, dispatch the consultant.
- Consultant scans its own subtree only.
- Reviewers / auditors scan consultant subtrees per their cards.
Boundaries
- Tidy only. Reorganize existing recorded content; never source-walk, never run MadGraph, never author a new principle. Generalization and re-verification are not this skill's job (see the header).
- Single-writer-per-page. Each owner tidies its own subtree; the lead tidies
lead/. Cross-subtree refs are flagged, not edited.
- Cheap by construction. Reorganizing already-cited content needs no source access; invoking the pass is the go-ahead for these fixes — no further asks.
