CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

calibrate

Re-tunes the vSR plugin's thresholds from the user's own decision log. Tunes only the parts the user names (for example just the routing difficulty lines), asks about the decisions closest to each line in plain words, fits new lines to the answers (moving each at most 0.15 per run), explains the change and recommends whether to apply it, and applies only on approval. Use when vSR asks for a calibration, when vSR holds, hides, warns or escalates too often or too rarely, when its subagent routing feels too weak or too strong, or after a change of classifier model.

Quality

95%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

Quality

Content

100%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.

The body is an exemplary procedural skill: an 8-step calibration workflow with per-step exit criteria, exact commands and question wording, bounded changes, approval gating and backup before writing settings, and feedback loops for short data. Detail is appropriately pushed to the script and one clearly signaled reference, with no padding or background explanation anywhere.

DimensionReasoningScore

Conciseness

The body is lean operational instruction with zero concept explanation — every sentence is skill-specific detail Claude could not infer (e.g. "a held command they declined counts as 'should ask', and pushback within 15 minutes of a Haiku or Sonnet route counts as 'needed stronger'"). It assumes competence and every token earns its place, matching anchor 5; anchor 4's 'minor instances of over-explanation' does not apply since no passage can be trimmed without losing a real constraint.

5 / 5

Actionability

Guidance is fully executable: exact commands with defined placeholders ("`<apps>` is the comma list chosen in step 1, for example `router`"; "`label --id <id> <decision id>=<option value> ...`"), exact AskUserQuestion wording ("header 'Tune what', question 'Which parts of vSR should I tune?'", "multiSelect: true"), and expected output markers to act on ("saved N label(s)", "wrote"). Per the code-vs-instruction scoring note, this instruction-only skill's guidance is copy-paste actionable, matching anchor 5.

5 / 5

Workflow Clarity

Eight sequenced steps each carry an explicit "Done when" checkpoint, with early-stop conditions ("When the log holds no decisions... stop"), feedback loops (step 4's answer-count check offering another sample round; step 8's "Answer more questions (back to step 3)"), and approval-gated apply with backup ("saves the previous copy beside the file"). The settings-writing batch operation is fully validated, so the batch-operation cap does not apply; this matches anchor 5's explicit validation, error-recovery loops, and per-step checklist.

5 / 5

Progressive Disclosure

The ~40-line body is a clean overview plus a "## Steps" section, with implementation detail correctly kept in scripts/calibrate.py (invoked, not inlined) and threshold meanings deferred to a single one-level-deep, well-signaled reference ("read [../setup/thresholds.md](../setup/thresholds.md) and answer in plain words"). No nested references and nothing inlined that belongs in a separate file; the under-50-line well-organized exception applies, matching anchor 5.

5 / 5

Total

20

/

20

Passed

Description

92%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 strong description that concretely enumerates the skill's entire behavior (scope restriction, plain-word questioning, bounded fitting, explain-recommend-apply gating) and pairs it with an explicit, symptom-oriented 'Use when' clause using natural both-direction phrasings. The only gap is that a few natural trigger terms for the check app are absent from the trigger list.

DimensionReasoningScore

Specificity

The description enumerates the full concrete behavior chain — "Tunes only the parts the user names", "asks about the decisions closest to each line in plain words", "fits new lines to the answers (moving each at most 0.15 per run)", "explains the change and recommends whether to apply it, and applies only on approval" — with a precise numeric constraint. This matches the anchor for multiple specific concrete actions with comprehensive coverage; score 4's 'minor gaps in coverage' does not apply since every phase of the skill is named.

5 / 5

Completeness

Both questions are answered explicitly: the 'what' is the full action list ("Re-tunes the vSR plugin's thresholds from the user's own decision log..."), and the 'when' is a literal "Use when..." clause with concrete trigger phrases. This is a direct match for anchor 5; anchor 4's 'when could be more explicit' does not apply since the triggers are enumerated.

5 / 5

Trigger Term Quality

Triggers include natural symptom phrasings with both-direction synonyms — "holds, hides, warns or escalates too often or too rarely", "subagent routing feels too weak or too strong", "after a change of classifier model" — which is good keyword coverage. It falls short of anchor 5's comprehensive coverage: common user phrasings for the check app (e.g. answers that overclaim) and terms like pushback/false alarms are not represented as triggers.

4 / 5

Distinctiveness Conflict Risk

It names a clear niche — calibrating the vSR plugin from its own decision log — with triggers tied to vSR-specific behavior ("when vSR asks for a calibration", "its subagent routing"), so no other skill would plausibly fire on these phrases. Minimal conflict risk matches anchor 5; the voice is also correctly third person ("Re-tunes", "Tunes", "asks").

5 / 5

Total

19

/

20

Passed

Validation

87%

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

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

allowed_tools_field

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

Warning

relative_links

Relative link issues: 1 suspicious

Warning

Total

14

/

16

Passed

Repository

vllm-project/semantic-router

Path

src/agent-harness/claude-code/skills/calibrate/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.