CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

correction-pipeline

When the user corrects a factual error, root-cause it immediately. Don't just note the correction — trace the error to its source, fix the source, and prevent recurrence. Every factual error is either a data error (bad brain page, bad memory file, bad rendered SOUL/USER identity, bad facts row) or a hallucination (LLM confabulated from partial signals).

Quality

82%

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

SKILL.md
Quality
Evals
Security

Quality

Content

88%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.

Excellent operational content: an explicitly ordered 8-step pipeline with per-step decision questions, copy-paste-ready gbrain/grep commands for every error class, validation and feedback loops for a mutating workflow, and disciplined delegation to sibling skills. The only weaknesses are mild redundancy between the classification table, fix steps, and contract section, and reference tables inlined in a long body.

DimensionReasoningScore

Conciseness

The body is dense and mostly efficient — the gbrain commands and error codes are custom-tool knowledge Claude doesn't have, so their inclusion is earned. Minor trims exist: the Step 5 "Fix surface" column restates Step 6, and the "Contract" section reiterates guarantees already covered by steps and anti-patterns, matching anchor 4's 'minor instances of over-explanation that could be trimmed'.

4 / 5

Actionability

Fully executable, copy-paste-ready commands with appropriate placeholders throughout: `gbrain search "<relevant terms>" --limit 10`, `gbrain forget <fact-id>`, `gbrain bootstrap interview --set KEY "corrected value"`, `grep -ri "<wrong claim terms>" "$BRAIN_DIR/people/"`, plus a concrete Step 8 report template and specific error-code recovery steps — matching anchor 5.

5 / 5

Workflow Clarity

"Run these steps IN ORDER. Report findings to the user" gives explicit sequencing, each step carries a decision question ("Is the wrong fact IN the brain?"), and despite being mutating/destructive, validation and feedback loops are present: bootstrap read-back with confirm hash, re-running `recall` instead of guessing fact ids, re-sync after repo edits, a Step 7 propagation check, and a "When it fails" section with per-error recovery — anchor 5.

5 / 5

Progressive Disclosure

Sections are well-organized and delegation is clearly signaled one level deep ("The full interview discipline... lives in skills/soul-audit/SKILL.md — route through it"; brain-first and agent-operator conventions linked with what they cover). No bundle files exist to verify, and the ~260-line body inlines tables (Severity Tiers, Recurring Error Patterns) that could sit in a reference file — anchor 4 rather than 5's fully split structure.

4 / 5

Total

18

/

20

Passed

Description

71%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: it answers both what and when explicitly, lists concrete actions and fix surfaces, and defines a distinctive niche. Its main gap is that the natural trigger phrases users would say are absent from the description itself, having been placed in sibling frontmatter fields, which limits trigger-term coverage.

Suggestions

Fold one or two quoted user trigger phrases into the description itself (e.g., "Use when the user says 'that's wrong' or 'that's not true'") so trigger-term coverage doesn't depend on the separate when_to_use field.

Name the core tool flow briefly in the description (e.g., search the brain, check memory/identity files, expire bad facts rows) to lift specificity from 'several actions' to comprehensive coverage.

Consider mentioning that proactive conflict detection is out of scope to further sharpen distinctiveness against the contradictions-probe skill.

DimensionReasoningScore

Specificity

"trace the error to its source, fix the source, and prevent recurrence" lists several concrete actions, and the enumeration of concrete surfaces ("bad brain page, bad memory file, bad rendered SOUL/USER identity, bad facts row") pins the domain. It falls short of anchor 5's comprehensive, tool-level coverage (no commands or fix procedures) but exceeds anchor 3's '1-2 concrete actions'.

4 / 5

Completeness

Both halves are present: what ("root-cause it immediately... fix the source, and prevent recurrence") and when ("When the user corrects a factual error"). No 'Use when' cap applies, but the when-clause lacks the concrete trigger phrases that anchor 5 requires, fitting anchor 4's "'when' could be more explicit or specific".

4 / 5

Trigger Term Quality

"When the user corrects a factual error" is the single natural scenario phrase; the concrete trigger phrases users actually say ("that's wrong", "that's not true") live in the separate when_to_use/triggers fields rather than the description, leaving common variations and synonyms missing — a match for anchor 3 rather than anchor 4's good coverage.

3 / 5

Distinctiveness Conflict Risk

The description carves a clear niche — correcting factual errors across brain pages, facts rows, and rendered SOUL/USER identity files — with triggers tied to user corrections, so it is clearly distinguishable from adjacent skills (proactive maintenance, citation formatting). Anchor 5's 'clear niche with distinct triggers; minimal conflict risk' fits best.

5 / 5

Total

16

/

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

frontmatter_unknown_keys

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

Warning

relative_links

Relative link issues: 3 suspicious

Warning

Total

14

/

16

Passed

Repository

garrytan/gbrain

Path

plugin-variants/gbrain-coding/skills/correction-pipeline/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.