dayan-agent-skills

Core knowledge map

This page is the public map of the ideas behind the Skills. It is intentionally a distilled map, not an export of any private knowledge base, customer work, operating memory, or internal playbook.

What belongs in the public library

Publish a method when it meets all four tests:

1. it solves a repeated problem, not a one-off story;
2. it can be used without private context;
3. it has a clear artifact or decision shape;
4. it has a verification path or an explicit human-review boundary.

Public method candidates

Method	Public shape	Best paired Skill	Evidence to add before promoting
Task contract first	Convert vague work into goal, context, boundary, authority, evidence, and stop signal	dayan-wenzhen	more accepted and rejected contract fixtures
Value test loop	Separate interesting output from a result someone would actually use, cite, adopt, or pay for	dayan-autoresearch-agent	public experiment templates and false-positive examples
False-positive control	Treat fluent output, pretty UI, and passing structure checks as insufficient evidence	dayan-adversarial-reviewer	promoted in docs/methods/false-positive-control.md with a public review fixture
Weak-signal radar	Turn scattered external signals into hypotheses, not conclusions	dayan-industry-research	source-level evidence packets and scoring rubrics
Project operating system	Run complex work as topology, wave, approval, evidence receipt, and archive	dayan-plan	sample task graph and recovery fixtures
Business evidence loop	Convert content, demos, and repository work into measurable external signals	dayan-ai-seo	adoption-report template with non-claim boundaries
Audience-clean delivery	Keep reader-facing artifacts free of production notes, placeholders, hidden instructions, and internal roles	dayan-content-production	more fixture pairs for clean vs rejected output
Tool-before-agent	Turn deterministic checks into scripts before asking an agent to remember them	dayan-hook-factory	promoted in docs/methods/tool-before-agent.md with a verifier-plan fixture
Human authority ledger	Name the actions that remain human-controlled before a workflow starts	dayan-agent-designer	promoted in docs/methods/human-authority-ledger.md with an authority-ledger fixture
Reusable asset pipeline	Package artifact, validator, evidence, handoff, version, and retirement rule together	dayan-marketplace-publishing	release examples where validators catch bad packages

Methods already published as cards

The current method-card set is in docs/methods.md. It covers complexity, decomposition, adversarial validation, evidence chains, reversible action, falsifiable framing, human authority, completion levels, project harnesses, delivery assets, simple communication, audience-clean artifacts, tool-before-agent, false-positive control, and human authority ledgers.

What not to publish

• personal memory;
• customer-specific material;
• credentials, private paths, private infrastructure, or account workflows;
• claims about revenue, adoption, ranking, or compatibility without public evidence;
• internal names that do not help an outside user apply the method.

Next public upgrades

1. Add one fixture for each method candidate.
2. Add one rejected or boundary fixture for each high-risk method.
3. Promote a candidate to a full method card only when the fixture and verifier exist.
4. Link the promoted method to one Skill, one artifact shape, and one verification command.

This site is open source. Improve this page.