Antipattern Detection Skill

This skill automatically activates after linting failures, test failures, or code review feedback to identify recurring antipatterns. It stores findings in ~/.claude/config/knowledge_base.yml (the YAML source of truth) via learning_capture.sh. The human-readable docs/KNOWLEDGE_BASE.md is auto-generated from the YAML by learning_capture.sh sync-docs.

Trigger Criteria

Activate when any of the following are detected in the current session:

Trigger	Detection Method
Lint failure	Exit code != 0 from ruff, eslint, golangci-lint, tflint
Test failure	Exit code != 0 from pytest, go test, vitest, terratest
Security finding	Any finding from bandit, gosec, npm audit, tfsec
Repeated pattern	Same issue type seen 3+ times across files
Code review feedback	Parallel agent consensus flags a recurring concern

Analysis Process

Step 1: Collect Failure Data

Gather from the current session:

•Linter output (error codes, rule names, affected lines)
•Test output (failure messages, assertion errors)
•Security scan output (vulnerability types, severity)
•Agent review feedback (concerns raised by 2+ agents)

Step 2: Pattern Classification

Classify each failure into an antipattern category:

Category	Examples
`security`	Hardcoded secrets, SQL injection, unsafe deserialization
`error-handling`	Bare exceptions, silent failures, missing error boundaries
`performance`	N+1 queries, unbounded loops, missing pagination
`type-safety`	Missing type hints, `any` overuse, unchecked casts
`testing`	Missing edge cases, brittle assertions, test pollution
`architecture`	Circular imports, god classes, tight coupling
`naming`	Misleading names, inconsistent conventions, abbreviations
`duplication`	Copy-paste code, reimplemented stdlib, redundant logic

Step 3: Deduplication

Before adding a new antipattern, check for existing entries in the knowledge base:

bash

~/.claude/scripts/learning_capture.sh query --category antipattern --language <LANG> --format llm

•
If a matching entry exists (same category + language + similar description), increment it:
bash
```
~/.claude/scripts/learning_capture.sh increment <ID>
```
•
Only create a new entry if no similar pattern is documented

Step 4: Store Entry

For each new antipattern, store it in the YAML knowledge base:

bash

~/.claude/scripts/learning_capture.sh add \
  --category antipattern \
  --language <language> \
  --title "<short title>" \
  --description "<1-2 sentence problem description and recommended fix>" \
  --tags "<comma-separated tags>" \
  --confidence <high|medium|low> \
  --source antipattern-detect

After storing all new entries, regenerate the human-readable docs:

bash

~/.claude/scripts/learning_capture.sh sync-docs

Storage

Source of truth: ~/.claude/config/knowledge_base.yml (machine-readable YAML)

Entries are stored via learning_capture.sh add which handles:

•Automatic ID generation (KB-NNN format)
•YAML validation before writing
•Header comment preservation

Human-readable view: docs/KNOWLEDGE_BASE.md (auto-generated)

Regenerated by learning_capture.sh sync-docs. Do not edit this file directly.

Non-Blocking Behavior

This skill follows the same non-blocking pattern as code-quality:

•Never blocks user workflow or command execution
•Reports inline when an antipattern is detected
•Writes to knowledge_base.yml via learning_capture.sh after primary task completes
•Suggests fixes but does not auto-apply changes

Integration

This skill complements other skills:

Skill	Relationship
`code-quality`	Feeds security/quality findings into antipattern detection
`learning-loop`	Shares knowledge_base.yml as the common data store
`verify`	Lint/test failures trigger antipattern analysis

When both code-quality and antipattern-detect trigger:

•code-quality provides immediate inline feedback
•antipattern-detect documents the pattern for future reference
•Results are complementary, not duplicated

Safety Checks

•Never modify source code -- only writes to knowledge_base.yml via learning_capture.sh
•learning_capture.sh validates YAML before writing
•Cap entries at 500 per knowledge base file (learning_capture.sh limit)
•Sanitize code examples to remove actual secrets or credentials
•sync-docs regenerates docs/KNOWLEDGE_BASE.md -- do not edit that file directly