skills/core/continuous-learning/SKILL.md
Extract reusable instincts from Claude Code sessions via hooks, score them by confidence, and evolve them into skills, commands, or agents.
npx skillsauth add bereniketech/claude_kit continuous-learningInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Use PreToolUse and PostToolUse hooks — never rely on skills to observe. Hooks fire 100% of the time; skills fire only 50–80%.
Register observe.sh for both hook events in ~/.claude/settings.json:
{
"hooks": {
"PreToolUse": [{ "matcher": "*", "hooks": [{ "type": "command", "command": "~/.claude/skills/continuous-learning-v2/hooks/observe.sh" }] }],
"PostToolUse": [{ "matcher": "*", "hooks": [{ "type": "command", "command": "~/.claude/skills/continuous-learning-v2/hooks/observe.sh" }] }]
}
}
Rule: Never use a Stop hook as the sole observation point — it misses mid-session patterns.
Detect project context in this priority order:
CLAUDE_PROJECT_DIR env var (highest priority)git remote get-url origin — hashed to 12 chars, portable across machinesgit rev-parse --show-toplevel — fallback, machine-specificStore project instincts under ~/.claude/homunculus/projects/<hash>/instincts/, global instincts under ~/.claude/homunculus/instincts/personal/.
Rule: Default new instincts to project scope; only promote to global when seen in 2+ projects.
Move instincts through four stages: observe → extract → validate → promote.
| Stage | Action | |---|---| | Observe | Hook captures tool calls, prompts, outcomes, project context | | Extract | Background observer (Haiku) detects user corrections, repeated patterns, error resolutions | | Validate | Confidence rises with repeated confirmation; falls with user corrections | | Promote | Project instinct with avg confidence ≥ 0.8 in 2+ projects → promote to global |
Each instinct is atomic: one trigger, one action, one domain tag (code-style, testing, git, debugging, workflow).
Score every instinct on a 0.3–0.9 scale:
| Score | Meaning | Behavior | |---|---|---| | 0.3 | Tentative | Suggested, not enforced | | 0.5 | Moderate | Applied when relevant | | 0.7 | Strong | Auto-approved | | 0.9 | Near-certain | Core behavior |
Increase confidence when the pattern recurs without correction. Decrease when the user explicitly contradicts it or it goes unobserved for extended periods.
Rule: Never hard-code an instinct at 0.9 from first observation — require at least 3 confirming occurrences.
| Pattern Type | Scope | |---|---| | Language/framework conventions, file structure, code style, error handling | project | | Security practices, general best practices, git conventions, tool workflow | global |
/instinct-status — Show all instincts (project + global) with confidence scores
/evolve — Cluster related instincts into skills, commands, or agents
/instinct-export — Export instincts (filterable by scope/domain)
/instinct-import — Import instincts from a file
/promote [id] — Promote a project instinct to global scope
/projects — List all known projects and their instinct counts
Run /evolve after accumulating 10+ instincts in a domain to cluster them into a reusable skill.
~/.claude/homunculus/
projects.json # Registry: hash → name/path/remote
instincts/personal/ # Global auto-learned instincts
instincts/inherited/ # Global imported instincts
evolved/{agents,skills,commands}/ # Global evolved artifacts
projects/<hash>/
project.json
observations.jsonl
instincts/personal/ # Project-scoped instincts
evolved/{skills,commands,agents}/
Rule: Never store raw conversation content in instincts — store only the extracted pattern.
testing
AUTHORIZED USE ONLY: This skill contains dual-use security techniques. Before proceeding with any bypass or analysis: > 1.
testing
Provide comprehensive techniques for attacking Microsoft Active Directory environments. Covers reconnaissance, credential harvesting, Kerberos attacks, lateral movement, privilege escalation, and domain dominance for red team operations and penetration testing.
development
Detects missing zeroization of sensitive data in source code and identifies zeroization removed by compiler optimizations, with assembly-level analysis, and control-flow verification. Use for auditing C/C++/Rust code handling secrets, keys, passwords, or other sensitive data.
development
Comprehensive guide to auditing web content against WCAG 2.2 guidelines with actionable remediation strategies.