a collection of agent skills
This repository also configures its own agents. See what skillery tells them →
Agent Claude Code
Reviews a change through exactly one review lens (L1–L5) and reports prioritised findings; read-only. Use when a caller fans out a pre-PR review, one code-reviewer per lens (a full review is five dispatches, per meaningfy-code-review). Not for fixing code (implementer) or discussing findings.
Agent Claude Code
Shapes an EPIC from human seeds and derives its clarity-gated PLAN (≥9/10) as an OpenSpec change, returning question rounds to the caller until no decision is open. Use when starting or refining an epic. Not for archiving or grooming living specs (spec-stewardship) or for writing code (implementer).
Agent Claude Code
Implements one ready task of an OpenSpec change as tested production code, test-first. Use when a task in tasks.md is ready (PLAN gated ≥9/10, Gherkin written). Not for planning (epic-planner), reviewing (code-reviewer) or committing.
Agent Claude Code
Builds and evolves a domain's conceptual model with the user, model-driven and data-centric, choosing the model source and rendering. Use when modelling a domain or evolving an ontology. Not for delegation; not for system architecture (system-architect).
Agent Claude Code
Writes a sendable proposal or offer email from client requirements, asking the questions it needs through its output, with estimate and executive framing. Use when a client asks for an offer. Not for pricing strategy alone (estimation).
Agent Claude Code
Designs a new system with the user before code: use cases, architectural decisions and system diagrams, stating its view and plan briefly before each step. Use when starting or reshaping a system's architecture. Not for delegation; not for code (implementer).
Agent Claude Code
Operates Meaningfy's infrastructure with a person who is not a devops engineer: reads both infra repositories and monitoring, plans each change, gets it reviewed, asks consent with a plain-words card, applies exactly what was shown, verifies and records it. Use when someone opens a session in the stacks repository to…
Agent Claude Code
Reviews one infrastructure change — an OpenSpec change's plan or a PR in either infra repository — against the devops skills and reports findings, blast radius, undo path and stack conformance; read-only. Use when a caller wants the four-eyes review before executing a plan, mid-implementation or before merging. Not…
Agent Claude Code
Watches Meaningfy's infrastructure with no human present: runs the daily check-up, answers alerts with allow-listed operations only when the unattended-action policy says act, parks everything else with a decision card, and reports every run to Slack and GitHub issues. Use when the schedule or an alert fires, or when…
Agent Claude Code
Keeps Meaningfy's Odoo sales pipelines true and moving: grooms deals and contacts with logged, reversible writes, prepares meetings and drafts, and asks the owner only through decision cards. Use when the scheduled open-day, close-day and weekly runs fire, or when the owner opens a sales working or intake session. Not…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: