CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

siege

Work on one hard mathematics problem in rounds: each round a judge writes a few self-contained questions, independent workers answer them, and the judge keeps a ledger of what is proved, refuted and open, until the judge concludes or the rounds run out, and then a proof.md says plainly what is and is not proved. A run takes hours and dozens of worker runs. Usage: /math-proof:siege [NAME=value settings] <the problem, stated in full, or the path of a file holding it>.

Quality

74%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./plugins/math-proof/skills/siege/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

73%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

An exceptionally actionable, well-validated orchestration protocol whose steps, commands, prompts and failure paths are copy-paste concrete. Its two weaknesses are structural: ~700 lines all inline in SKILL.md where the brief templates belong in separate reference files, and enough repeated restatement of the same rules that a tighter document could carry the same protocol in far fewer tokens.

Suggestions

Move the four brief templates (PLAN, DRAFT, REFINE, SELECT, FINALIZE) and their slot texts ({CONCLUDE_RULE}, {ESC_SUMMARY}, {ESC_WAVE}, {PARTIALS}, {PARTIALS_FIN}) into references/ files (e.g. references/plan-brief.md), leaving SKILL.md to state when each brief is launched and how its slots are filled.

State each rule once and reference it by name elsewhere — the ESC_ROUND worker-type and wave-cap rules are currently given in the approach section, in Settings, in Rules, and again inside the slot texts, and the 'do not copy the problem statement into query files' instruction appears in at least four places.

Break the 150-250-word legalistic run-on sentences (e.g. the four-paragraph Settings error-handling instruction and PLAN BRIEF item 2 (ii)) into shorter sentences or bullet lists; the content is procedural, not contractual, and would lose no precision.

DimensionReasoningScore

Conciseness

The 733-line body contains no explanation of concepts Claude already knows — nearly every sentence is procedural — but the legalistic run-on style restates the same rules repeatedly (the ESC_ROUND worker-type and wave-cap rules appear in the approach section, in Settings, in Rules, and again inside the {ESC_SUMMARY}/{ESC_WAVE} slot texts). Fits anchor 3 ('mostly efficient but could be tightened'): the excess is redundancy, not explanation of known ground, so it stays above anchor 2.

3 / 5

Actionability

Fully executable throughout: exact commands ("SCRIPT check DIR {r} {CAP} {MIN_ROUNDS} {attempt}"), a verbatim worker prompt template with the exact end line "=== END OF ANSWER {Q} ===", the fallback chain for locating ledger.py, prescribed actions for every verdict line (WAVE/CONCLUDE/RETRY/TAIL), and the literal grep command for the standalone check. Copy-paste ready with specific handling for the common and error cases — anchor 5.

5 / 5

Workflow Clarity

The round loop (a)-(d) and proof tail (e)-(i) are explicitly sequenced with a mechanical validation layer throughout: the SCRIPT check verdicts, RETRY feedback loops, the SCRIPT answers finished/partial decision, the FLOOR stop rule, the SCRIPT gate short/full-tail decision, and the final grep standalone check with a repair pass. Explicit validation steps, feedback loops, and error recovery match anchor 5 exactly.

5 / 5

Progressive Disclosure

A 733-line monolith: the PLAN BRIEF with its {CONCLUDE_RULE}/{HORIZON}/{ESC_SUMMARY}/{ESC_WAVE} slot texts alone runs roughly 300 lines, all inlined in SKILL.md, where these brief templates clearly belong in separate files loaded at need. The one external file (scripts/ledger.py) is clearly signaled, but there is no reference structure at all — fits anchor 2 ('content that clearly belongs in separate files is inlined'); not anchor 1 because section headers give real structure, not anchor 3 because the inline briefs go far beyond 'could be better organized'.

2 / 5

Total

15

/

20

Passed

Description

66%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A specific, distinctive description of an unusual orchestration protocol, weakened only by the absence of any 'use when' trigger guidance and by missing the most natural user phrasings (prove, theorem, conjecture). The 'what' is strong; the 'when' is entirely implicit.

Suggestions

Add an explicit trigger clause, e.g. "Use when the user poses a hard mathematics problem to prove or disprove — a conjecture, an Olympiad-style statement, or a claimed proof to verify — and a single-pass answer is unlikely to settle it."

Include the natural verb forms and synonyms users would actually say — "prove", "disprove", "theorem", "conjecture", "verify this proof" — alongside the existing "proved/refuted" terms.

Mention the resume capability (re-running the same command in the same directory continues a stopped run), since it is a key behavior a user would want to discover from the description.

DimensionReasoningScore

Specificity

Lists several concrete, distinct actions — "a judge writes a few self-contained questions, independent workers answer them", "keeps a ledger of what is proved, refuted and open", "a proof.md says plainly what is and is not proved" — but capabilities like resume behavior and the settings system are not covered, so it stops short of comprehensive. It clearly exceeds the 1-2-concrete-actions level of anchor 3.

4 / 5

Completeness

The 'what' is clear and detailed (rounds of judge/worker queries, a claims ledger, a proof.md deliverable), but there is no 'Use when...' clause or equivalent trigger guidance — the "Usage:" line is command syntax, not a when-to-invoke condition. Per the judging guidelines, a missing explicit trigger clause caps completeness at 3.

3 / 5

Trigger Term Quality

"hard mathematics problem", "proved", "refuted" and "proof.md" give good natural keyword coverage for a user asking to attack or verify a hard claim, but the most common user phrasings ("prove", "disprove", "theorem", "conjecture") are missing. Coverage is a notch above anchor 3's 'some relevant keywords' level, but short of anchor 5's synonym-and-extension completeness.

4 / 5

Distinctiveness Conflict Risk

The multi-round judge/worker/ledger siege mechanism is a highly distinctive niche unlikely to trigger the wrong skill, and only the broad "hard mathematics problem" trigger domain overlaps with general math-assistance skills. Fits anchor 4 (mostly distinct, minor overlap risk) better than anchor 5's 'minimal conflict risk'.

4 / 5

Total

15

/

20

Passed

Validation

81%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 13 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

skill_md_line_count

SKILL.md is long (733 lines); consider splitting into references/ and linking

Warning

allowed_tools_field

'allowed-tools' contains unusual tool name(s)

Warning

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

13

/

16

Passed

Repository

anthropics/claude-plugins-official

Path

plugins/math-proof/skills/siege/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.