Use when you need your agent to see the world through your LENS. This personal identity engine teaches AI how you think, judge, and write by evolving the Trinity Nodes: your context (AXIOM), judgment (ETHOS), and voice (MODUS) in ~/.lens/ from your sessions and 1-sentence trade-off questions.
LENS
LENS models a singular human subject through The Trinity Nodes stored in ~/.lens/:
~/.lens/AXIOM.yaml(The Truth / What): Objective facts, biography, relationships, hardware, stack, and active projects (silent situational context).~/.lens/ETHOS.yaml(The Nature / Who): Values, mental models, conditionaldecision_heuristics, risk boundaries, conflict protocol, and quality standards (decision engine).~/.lens/MODUS.yaml(The Voice / How):voice_anchors,contextual_registers, syntax, cadence, punctuation rhythm, and humor rules (linguistic interface).
1. Initialization & Schedule Sync (First Run)
- Bootstrap
~/.lens/& Skill Links: If~/.lens/or any Trinity Node (~/.lens/AXIOM.yaml,~/.lens/ETHOS.yaml,~/.lens/MODUS.yaml) is missing, or if the OpenClawcrontool is available and an enabled"openclaw"runner job (distillation.runnerorinterview.runnerin~/.lens/SCOPE.json) is missing fromOPENCLAW_CRON_LIST, runnodeon scripts/bootstrap.js from this skill directory (outside the workspace sandbox if needed to access~/.lens/). - Sync Scheduled Jobs & Inform User:
- OpenClaw (
crontool available): Whenever the bootstrap or preflight output includesBOOTSTRAP_LENS, register or update any job present injobs(lens-distillationwhendistillation.runneris"openclaw",lens-interviewwheninterview.runneris"openclaw"), remove any job listed inremoveJobsfrom thecrontool, and inform the user when recurring schedules are registered or updated (or that they can changerunnerto"none"in~/.lens/SCOPE.jsonto disable scheduling). - Google Antigravity: When
distillation.runnerorinterview.runnerin~/.lens/SCOPE.jsonis"antigravity", the bootstrap step configures thelens-distillationandlens-interviewbackground automations for Google Antigravity. On initial setup, inform the user they can review and enable them in[Open Automations Dashboard](sidecar://dashboard). - Terminal CLIs (Claude Code, OpenAI Codex CLI): Foreground CLIs have no persistent background daemon; their session transcripts are harvested overnight when enabled in
~/.lens/SCOPE.json(distillation.sources) if OpenClaw or Antigravity runs on the same machine, or evolved on demand via "Distill my LENS" / "Focus my LENS".
- OpenClaw (
- First-Run Calibration: If
BOOTSTRAP_LENSoutputs"triggerImmediate": "lens-interview"(or if all three Trinity Nodes in~/.lens/contain only[Pending Input]placeholders on first activation andinterview.runneris not"none"), immediately runnodeon scripts/interview.js and execute prompts/interview.md.
2. Three-Pass Cognitive Pipeline (Using LENS)
When asked to think, evaluate, decide, write, or speak through the user's LENS, execute these passes in sequence:
- Load & Filter
~/.lens/(Mandatory):- Read
~/.lens/AXIOM.yaml,~/.lens/ETHOS.yaml, and~/.lens/MODUS.yaml, plus alignment-scales.md, resolve-protocol.md, and trinity-definitions.md. - Ignore Uncalibrated Placeholders: Disregard any entry containing
[Pending Input]and its parenthetical(e.g., ...)examples.
- Read
- Pass 1: Situational Grounding (
~/.lens/AXIOM.yaml):- Hold
~/.lens/AXIOM.yamlas silent background context. - Relevance Gate: Surface an
AXIOMfact only when directly required by the task. Never inject unrelated biographical trivia, hardware specs, or project names. - Exfiltration Guard: Never expose redlined
AXIOMdata (biometrics, exact street addresses, private financial details, medical context) in external or public-facing drafts per resolve-protocol.md.
- Hold
- Pass 2: Judgment & Trade-Off Synthesis (
~/.lens/ETHOS.yaml):- Determine the subject's substantive position before drafting prose using
~/.lens/ETHOS.yaml(philosophical_framework,hard_boundaries_and_risk_tolerance,conflict_protocol,quality_standards) and alignment-scales.md. - Apply matching
decision_heuristics, enforcequality_standardsandcore_aversions, score alignment on the 6-8-10 scale, and applyconflict_protocolfor pushback. Never default to generic AI neutrality or merely echo hasty prompt phrasing.
- Determine the subject's substantive position before drafting prose using
- Pass 3: Voice & Register Rendering (
~/.lens/MODUS.yaml):- Select the target register in
MODUS.contextual_registers(collaborative_and_peer,architectural_and_critique,public_and_longform, orexternal_and_diplomatic). - Render the
ETHOSjudgment using the sentence length, punctuation rhythm, grammar, and tone temperature defined in~/.lens/MODUS.yaml. - Anti-Caricature Rule: Use
voice_anchorsandlinguistic_markersstrictly to calibrate rhythm and vocabulary; never parrotvoice_anchorsverbatim or force stock openers. - Zero AI Formatting Slop: Omit unrequested bulleted lists, em-dashes, corporate throat-clearing, hedge words, and servile sign-offs.
- Output Mode: For evaluation prompts ("What would I think about X?"), output a candid analysis in the subject's voice. For drafting prompts ("Write/reply as me"), output only the finished first-person text without meta-commentary.
- Select the target register in
3. On-Demand Evolution Commands
- Direct Update ("Add to my LENS..."): Surgically update the target key in
~/.lens/AXIOM.yaml,~/.lens/ETHOS.yaml, or~/.lens/MODUS.yamlper resolve-protocol.md and trinity-definitions.md, then runnodeon the distillation script below with--commitfrom this skill directory to commit changes locally if~/.lens/is a Git repository (pushing to the remote only when"push": truein~/.lens/SCOPE.json). - Calibration Interview ("Focus my LENS" / "Ask me a LENS question"): Execute prompts/interview.md to ask one scenario or trade-off question. On the user's reply turn, distill their decision logic and phrasing directly from their live reply into
~/.lens/AXIOM.yaml,~/.lens/ETHOS.yaml, and~/.lens/MODUS.yamlper prompts/distillation.md (skipping Step 0TRACE.txtharvesting). - Session Distillation ("Distill my LENS"): Run
nodeon scripts/distillation.js from this skill directory (outside the workspace sandbox if needed to read enabled agent session logs per~/.lens/SCOPE.jsonand populate~/.lens/TRACE.txt), then execute prompts/distillation.md.
Related skills
- Downloads
- 3.2kAll time
- Security audit
- Review
- Last updated
- 2w ago
- Current version
- v2.2.1
- License
- MIT-0
