/instinct — Pattern Lifecycle | pds | ClaudePluginHub

Skill

/instinct — Pattern Lifecycle

From pds

Records, reviews, and promotes engineering patterns as instincts that mature into skills. Use when spotting recurring patterns, in post-swarm analysis, or at promotion thresholds.

developer-tools

$

npx claudepluginhub rmzi/portable-dev-system --plugin pds

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Instincts are lightweight engineering patterns — observations that accumulate confidence and promote to skills when mature. They live in `.claude/instincts.md` as the project's persistent lexicon.

SKILL.md

Similar Skills

instinct-system

188

Tracks project-specific code patterns as confidence-scored instincts through observe-hypothesize-confirm cycles, storing in .claude/instincts.md for promotion to MEMORY.md.

dotnet-claude-kit

learn

40

Extracts reusable code style, workflow, architecture, and tool patterns from git history, diffs, and observations into confidence-scored instincts. Run after features, bugs, or sessions.

atv-starter-kit

continuous-learning-v2

13

Observes Claude Code sessions via hooks to build atomic instincts with confidence scoring, evolves them into skills/commands/agents, adds project-scoped storage to isolate learnings.

8 files

everything-claude-code

Stats

Stars19

Forks0

Last CommitFeb 27, 2026

Actions

View Source View Plugin View on GitHub View README

Tags

instinct-lifecycle

engineering-patterns

skill-promotion

claude-instincts

Help us improve

Share bugs, ideas, or general feedback.

/instinct — Pattern Lifecycle

Instincts are lightweight engineering patterns — observations that accumulate confidence and promote to skills when mature. They live in .claude/instincts.md as the project's persistent lexicon.

Record an Instinct

When you observe a recurring pattern during work:

Read .claude/instincts.md
Check if the pattern already exists — if so, bump Times seen and adjust Confidence
If new, append an entry under ## Instincts:

### [descriptive title]
- **Observed**: [YYYY-MM-DD]
- **Times seen**: 1
- **Confidence**: low
- **Context**: [where/when — e.g., "during API integration swarms"]
- **Pattern**: [what happens — e.g., "workers duplicate auth setup across worktrees"]
- **Action**: [what to do — e.g., "extract shared auth config to a setup script"]
- **Status**: active

Confidence Levels

Level	Criteria	Next step
`low`	Seen 1 time, single context	Validate in future work
`medium`	Seen 2 times across different contexts	Continue tracking
`high`	Seen 3+ times, consistent pattern	Propose skill promotion

Validate Instincts

During post-swarm analysis (Phase 6) or /scout runs:

Read .claude/instincts.md
For each active instinct, check if it was re-observed
If yes: increment Times seen, upgrade Confidence if threshold met
If contradicted: mark Status: retired with a note
Propose new instincts for patterns not yet captured

Promote to Skill

When an instinct reaches high confidence:

Scout flags it in the meta-improvement report under ### Instincts
Scout drafts a skill file in skills/ (plan mode — human must approve)
After skill is merged, update the instinct: Status: promoted
Add a note: Promoted to /skill-name on YYYY-MM-DD

Retire an Instinct

Mark Status: retired when:

The pattern was wrong (contradicted by evidence)
The underlying cause was fixed (pattern no longer occurs)
The pattern is too context-specific to generalize

See Also

.claude/instincts.md — instinct storage
/pds:swarm — Phase 6 triggers instinct review
Scout agent — automates the lifecycle