docs(harness-principles): add 2 principles from 12-Factor Agents#28
Draft
markhazlett wants to merge 1 commit into
Draft
docs(harness-principles): add 2 principles from 12-Factor Agents#28markhazlett wants to merge 1 commit into
markhazlett wants to merge 1 commit into
Conversation
…anLayer) - Deterministic Routing, LLM Reasoning (§64) — use deterministic code to route between agent states; LLM for in-branch reasoning only - Context Window as Primary Execution State Store (§65) — infer execution metadata from event history, minimize auxiliary state - Add Sources Mined section (first entries: 12-factor-agents, Cognition, Anthropic engineering blog) - Update Part XII checklist with §64 and §65 items Primary source: https://github.com/humanlayer/12-factor-agents
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Mines Dex Horthy's 12-Factor Agents (HumanLayer, 2025) for two transferable principles that don't exist in the current 63-entry doc. Also bootstraps the Sources Mined section so future runs know what's already been covered.
Both principles clear the full quality bar: transferable, concrete mechanism, primary source verified via GitHub raw fetch, non-duplicative against existing §1–63, confidence high.
Principles Added
§64 — Deterministic Routing, LLM Reasoning (confidence: high)
The mechanism: classify input with a narrow LLM call, then route via explicit code to smaller, focused sub-prompts with fewer available tools. The doc's existing §29–34 cover phase-staging within a workflow; this is about the layer above — deciding which workflow to enter. Currently absent from the doc.
§65 — Context Window as the Primary Execution State Store (confidence: high)
The mechanism: execution metadata (current step, wait conditions, retry counts) is derived from the event log in context, not from a separate state table. Auxiliary state is minimized to items that cannot enter context (credentials, session IDs). Enables checkpointing and thread-forking as natural side effects. Complements §5 (layered context) and §39 (re-injection) without duplicating them.
Sources
Primary (verified via fetch):
Researched, primary source inaccessible (HTTP 403):
The 403-blocked sources are logged in Sources Mined for retry on the next run.
Not in scope
This PR only edits
.claude/docs/harness-principles.md. Skills, hooks, agents, and commands are unchanged. Adoption of §64 or §65 (e.g., a deterministic skill-dispatch layer) is a follow-up task if approved.Generated by Claude Code