From loom
Run Loom's compounding pass before closure. Use when accepted lessons should move into owner layers and memory needs support-only cleanup.
npx claudepluginhub z3z1ma/agent-loom --plugin loomThis skill uses the workspace's default tool permissions.
Retrospective is Loom's compounding workflow.
Guides capturing high-quality, generalizable learnings from ClosedLoop runs using decision tree, rejection criteria, and workflow to classify into CLAUDE.md or org-patterns.toon.
Analyzes conversations after significant work or 'reflect' triggers to extract learnings, classify them, and integrate into laws, skills, rules, or documentation via structured tasks.
Share bugs, ideas, or general feedback.
Retrospective is Loom's compounding workflow.
It assimilates what was learned during a ticket, initiative, critique, spike, debugging pass, or recent work slice into the existing owner layers.
It promotes learning into the records that already own that kind of knowledge. This skill owns retrospective workflow mechanics; shared record grammar only reminds operators that retrospective is a workflow route, not a new record kind.
loom-wikiloom-researchloom-specsloom-plansloom-initiativesloom-constitutionloom-evidenceloom-memory; not project truthUse the loop observe -> distill -> promote -> prevent.
Observe concrete signals from ticket journals, packet outputs, critique findings, evidence records, changed owner records, and recurring questions that had to be reconstructed.
Distill durable lessons separately from one-time execution detail, then promote each accepted lesson to the owner layer that can maintain it.
Prevent repeated mistakes by choosing exactly one prevention artifact when practical:
loom-memory after owner truth is
promoted or ruled out; not project truthIf there is nothing durable to promote, say so. Do not invent artifacts to look productive.
# Retrospective / Promotion Disposition as the standard closure home, with
# Wiki Disposition only for the wiki-specific outcome when wiki is one
promotion pathRead immediately for retrospective work:
skills/loom-records/references/retrospective.md when checking the shared
grammar that retrospective creates no new canonical record layer.skills/loom-tickets/SKILL.md when ticket acceptance, closure, or follow-up
disposition needs updating.Then read conditionally:
skills/loom-wiki/SKILL.md when accepted explanation should become reusable.skills/loom-research/SKILL.md when findings, rejected options, or null
results should remain citable.skills/loom-specs/SKILL.md, skills/loom-plans/SKILL.md,
skills/loom-initiatives/SKILL.md, or skills/loom-constitution/SKILL.md
when the lesson changes the owner truth those layers maintain.skills/loom-evidence/SKILL.md when observed artifacts, challenged
observations, or validation artifacts need preservation.skills/loom-memory/SKILL.md when support-only memory context may need
cleanup after owner truth is promoted or ruled out, replacement with
owner-record pointers, stale marking, or pruning.