← ClaudeAtlas

agentic-workflow-evolutionlisted

Improve a human-directs-AI-agents workflow through evidence rather than theory. Capture and route workflow observations; distinguish project-local lessons, cross-project candidates, and plugin-wide practices; design real-work experiments; evaluate reviewer, CI, migration, memory, gate, or session failures; promote, reshape, or retire practices; and reconcile the Issues, PRs, rules, practices, skills, and plugin releases affected by an approved change. Use when deciding whether a lesson belongs in pending-lessons, practices, a GitHub Issue, project rules, or a reusable plugin; when updating a skill; when a recurring failure suggests a workflow change; or when auditing whether current guidance is still supported. Triggers: "should we change the workflow", "promote this lesson", "update the plugin", "workflow experiment", "named failure mode", "why did this recur", "is this practice reusable", "retire the old practice", "log this revision", "what evidence supports this rule".
gidde032/agentic-workflow · ★ 0 · AI & Automation · score 63
Install: claude install-skill gidde032/agentic-workflow
# Agentic Workflow Evolution Treat the workflow as a versioned artifact. Improve it when evidence supports a specific mechanism, at the narrowest adoption level that solves the observed problem. Do not turn one project's preference into a universal rule, and do not leave a proven cross-project improvement trapped in session memory. The agent proposes and assembles evidence. The human decides whether project rules, gates, reusable skills, or plugin behavior change. Worked examples and provenance: `examples.md`. ## 1. Choose the adoption level first Do not ask only, "Is this a good idea?" Ask, "What scope does the evidence justify?" | Level | Meaning | Normal destination | |---|---|---| | Observation | Something happened once; the mechanism may be unclear | Session evidence or `pending-lessons.md` | | Project-local candidate | A plausible mechanism is worth testing in this repository | `pending-lessons.md`, experiment record, or project Issue | | Project-local adopted | Evidence supports an operating rule or structural protection here | Project rules, `practices/`, tests, or configuration | | Cross-project candidate | Similar evidence appears in materially different projects | Plugin backlog, explicit experiment, or proposal Issue | | Plugin-wide adopted | Evidence supports changing reusable skills or plugin behavior | Canonical plugin source and versioned release | | Retired | The practice is superseded, disproven, or no longer applicable | Preserved retirement record pl