Skip to content

inspiration.md

Latest commit

 

History

History
91 lines (71 loc) · 4.7 KB

inspiration.md

File metadata and controls

91 lines (71 loc) · 4.7 KB

Inspiration and Prior Art

AKC did not emerge in a vacuum. Several of its core design decisions are adapted from an earlier, more opinionated research repository:

Upstream: a daily signal filter

github.com/shimo4228/claude-skill-daily-research

Before Contemplative Agent existed, the author was running a daily research pipeline — a claude -p-driven cron job that selected two trending topics per day, conducted multi-stage web research, and wrote signal-filtered reports to an Obsidian vault. Two things about this upstream matter for AKC:

  1. It is how Contemplative Agent got started. One of the daily reports surfaced Laukkonen et al. (2025)'s Contemplative Inference paper. That report is what motivated the author to begin the Contemplative Agent project, whose engineering substrate in turn became the prior art for several of the ADRs listed below.
  2. It is how ADR-0010 got written. The LLM-Wiki-pattern report that triggered ADR-0010's conversation was itself a daily-research output. The pipeline that surfaced the question also implemented the answer — a fact noted in ADR-0010's "Origin loop" Notes.

The pipeline is therefore both the earliest implementation of AKC's Research phase (it predates the ADR by about a year) and the signal filter that made every downstream AKC decision possible. It was skillified in April 2026 under the claude-skill-daily-research name so it installs as a standard Claude Code skill. The pattern the pipeline embodies is documented, in abstract form and without personal references, in docs/skills/signal-first-research.md.

Contemplative Agent

github.com/shimo4228/contemplative-agent

Contemplative Agent is an independent research project exploring Contemplative Constitutional AI (Laukkonen et al., 2025) on a local 9B language model. It is an opinionated agent: it ships with a specific constitution, a specific identity template, and a specific research goal.

Underneath that opinionated surface is a set of engineering decisions that are useful regardless of the philosophy on top. AKC lifts those decisions into a harness-neutral form:

AKC component Prior art in contemplative-agent
ADR-0002 Immutable Episode Log src/contemplative_agent/core/episode_log.py, ADR-0004
ADR-0003 Three-Layer Distillation docs/adr/0004-three-layer-memory-architecture.md
ADR-0004 Two-Stage Distill Pipeline docs/adr/0008-two-stage-distill-pipeline.md
ADR-0005 Human Approval Gate docs/adr/0012-human-approval-gate-for-behavior-modifying-commands.md
examples/minimal_harness/ src/contemplative_agent/core/{episode_log,knowledge_store,distill}.py

What AKC does not carry over

AKC is explicitly not a contemplative framework. The following contemplative-agent elements are left upstream and are not part of AKC:

  • The four axioms (emptiness, non-duality, mindfulness, boundless care)
  • The identity templates in config/templates/contemplative/
  • The meditation adapter (adapters/meditation/) and Active Inference experiments
  • docs/adr/0002-paper-faithful-ccai-implementation.md
  • Any prompt text that references CCAI, axioms, or contemplative practice
  • The security triplet (Security by Absence, Single External Adapter, Untrusted Content Boundary) — these were carried in v1.x as ADR-0001, ADR-0006, and ADR-0007 but were extracted in v2.0.0 as genre-specific rather than cycle-defining. See the ADR-0009 addendum and ADR-0011 for the positioning rationale.

AKC takes the engineering substrate that is genre-neutral; contemplative-agent keeps the philosophy and the genre-specific security stance. Both are valid, and they are intentionally kept as separate repositories so that each can evolve independently.

Relationship model

Contemplative Agent is prior art and inspiration, not a downstream implementation. AKC does not depend on it; contemplative-agent does not depend on AKC. Changes in one repository do not require changes in the other. When AKC files are adapted from contemplative-agent source, the adapted files carry an in-file credit at the top.

Citing contemplative-agent

If your work builds on the engineering substrate (memory layers, distill pipeline) and you found it through AKC, please also credit the original repository:

Shimomoto, T. (2026). Contemplative Agent. https://github.com/shimo4228/contemplative-agent