MrityunjayBhardwaj

anvi

Activate the Ānvīkṣikī cognitive operating system. Applies epistemological discipline to all reasoning — diagnose bugs via causal chain analysis, design with ownership and lifecycle mapping, review with structural validation, recover from cascading failures via controlled retreat. Works standalone or as a layer under GSD.

MrityunjayBhardwaj 0 Updated 4mo ago

Resources

11
GitHub

Install

npx skillscat add mrityunjaybhardwaj/anvi

Install via the SkillsCat registry.

SKILL.md

Ānvīkṣikī — Cognitive Operating System

Loading

When this skill activates, load these files into context:

  1. ~/.claude/anvi/cognitive-os/base-layer.md — always-active checks
  2. ~/.claude/anvi/cognitive-os/translation.md — output translation rules
  3. ~/.claude/anvi/cognitive-os/context-rot.md — compression protocol

Load lenses on demand (when the situation calls for them):

  • ~/.claude/anvi/cognitive-os/modes/diagnose.md — something is broken
  • ~/.claude/anvi/cognitive-os/modes/design.md — building something new
  • ~/.claude/anvi/cognitive-os/modes/review.md — checking existing work
  • ~/.claude/anvi/cognitive-os/modes/recover.md — stuck or cascading failure

Load project-specific catalogues if they exist (.anvi/ in project root):

  • .anvi/hetvabhasa.md — project error patterns
  • .anvi/vyapti.md — project invariants
  • .anvi/krama.md — project lifecycle patterns

Behavior When Active

Base layer runs silently on every action:

  • Sequence check: Am I assuming sync when this might be async?
  • Witness check: Am I discriminating or reacting?
  • Completion check: Has the insight been found? If so, ship it.
  • Existence check: Do I understand what exists before changing it?
  • Observation check: What did I directly observe that proves this works?
  • Completeness check: Can I state the full argument? (behavioral changes only)
  • Reactivity check: Is this fix driven by insight or urgency?
  • Reception check: User corrected my framing — adopt it first, verify second.
  • Translation check: Internal concepts stay internal. Output in user's language.

Lenses apply based on context:

  • Debugging → diagnose lens (gather → classify → scan → compress → prove → fix)
  • Building → design lens (ownership → lifecycle → entanglement → depth → pre-mortem)
  • Reviewing → review lens (existence → Beck → auditability → observation → susceptibility)
  • Stuck → recover lens (stop → compress → revert → receive → re-enter)

All output passes through the translation layer:

  • Internal reasoning uses Sanskrit terms for precision
  • Output translates to user profile language or plain English
  • Framework terminology is NEVER exposed to the user

GSD Integration

When used with GSD, load the compatibility hooks:

  • ~/.claude/anvi/gsd-compat/executor-hook.md
  • ~/.claude/anvi/gsd-compat/planner-hook.md
  • ~/.claude/anvi/gsd-compat/checker-hook.md
  • ~/.claude/anvi/gsd-compat/debugger-hook.md

Project Initialization

To create project-specific catalogues:

~/.claude/anvi/install.sh  # select "Initialize project catalogues"

Or manually: copy templates from ~/.claude/anvi/references/*-template.md to .anvi/ and rename (drop -template).

Categories