Pattern Learn

Skill name: pattern-learn

Learn from normalized downstream failure events. Use after judge, PR review, GitHub review, CI, QA, or manual notes emit docs/issues/<branch>/learning-events/*.yaml to classify findings, create pattern rule proposals or rule modifications, and write non-rule learning recommendations.

Use this after downstream evaluators produce learning events. It consumes failure signals and proposes durable improvements. It does not change active rules.

Deterministic helper:

python3 /Users/trent.brown/agentic-development-workflow/scripts/pattern_tool.py event-inventory <issue-dir> <pattern-dir>

Workflow:

• Fail if no applicable .pattern-review directory is found after the normal scope walk or if an explicit pattern directory does not exist. Tell the caller to run /pattern-init <intended-scope>.
• Display the .pattern-review directory that will receive proposals or be checked for processed provenance.
• Read docs/issues/<branch>/learning-events/*.yaml. Each event represents one finding.
• Use event IDs and fingerprints to skip already-processed events by checking provenance in rules.yaml, proposals.yaml, deferred.yaml, and rejected.yaml.
• Classify each pending event as one of: new_pattern_rule, existing_rule_missed, existing_rule_ignored, process_gap, documentation_gap, test_gap, domain_knowledge, one_off_bug, or disputed_or_unclear.
• For rule-worthy outcomes, write new_rule or modify_rule proposals to .pattern-review/proposals.yaml with event provenance. For modifications, include a complete resolved rule candidate; do not patch active rules. Preserve evidence as structured source locations with path relative to the .pattern-review scope root and, when known, line. Avoid machine-specific absolute paths unless the source is outside any portable checkout context. Do not encode Markdown links in YAML; reports can render clickable links from those fields.
• For non-rule outcomes, write docs/issues/<branch>/learning-recommendations.md or the analogous source-event location.
• Normalize lifecycle YAML after writing proposals: python3 /Users/trent.brown/agentic-development-workflow/scripts/pattern_tool.py normalize-buckets <pattern-dir>.
• Parse the normalized lifecycle files before reporting success. Treat parse failures as command failures.
• Do not mutate learning-event source files. They are evidence, not state.
• Do not modify rules.yaml; only pattern-promote changes active rules.