Claude Skills by cavi-ai

Use when a spec-based build or tracker is complete and a retrospective is needed, or when closing out a build against its spec.

Use when running as a dispatched cloud Claude Code session against an Obsidian vault repo — do the task, write the result back as a reply note, and open a PR so Companion can pull it into the vault on any device.

Use when looking for non-obvious conceptual relationships between notes that are not linked yet. Read-only discovery that ranks and explains candidate connections; it proposes, it does not edit. To write the links, use wikilink-weaver.

Use when tagging notes, applying tags to new or untagged notes, or cleaning up tag sprawl and inconsistency in an Obsidian vault.

Use when summarizing or recapping recent vault activity — a daily, weekly, or periodic review of what changed, what was decided, and what is still open in a time window. For every open task across the whole vault with no time bound, use task-harvester.

Use when finding and merging duplicate or near-duplicate notes, consolidating notes that cover the same thing, or cleaning up redundant notes in an Obsidian vault.

Use when auditing or normalizing note frontmatter to a consistent schema across a folder or the vault. Defers to the vault's declared ontology types where they exist and proposes a schema only where they are silent.

Use when running any manifest-* advisor skill — the shared gather, prioritize, present, and route spine they all follow. Invoked by those skills, not directly by the user.

Use when asked to optimize, clean up, audit, or improve an Obsidian vault, assess vault health, or find structural problems in a vault.

Use when prioritizing or proposing work from vault notes — a roadmap, feature backlog, content plan, research agenda, risk register, or infra design, ranked and grounded in the notes. Pass the lens - pm (work already scoped), feature (new scope from feedback), content, infra, research, or risk.

Use when turning raw meeting notes, voice memos, or messy capture into a structured note with decisions, action items, and attendees.

Use when building or refreshing a Map of Content (MOC), creating an index or hub note for a topic or folder, or organizing related notes under one navigational note.

Use when a note covers too many topics and should be split into atomic notes, breaking up a bloated note, or extracting sections into their own linked notes.

Use when turning notes, data, or a plan into a beautiful, self-contained HTML artifact (report, dashboard, plan, table, or diagram) in the Anthropic-flavored "HTML is all you need" style — especially for Obsidian notes rendered in a ```claude-html``` block.

Use when expanding an outline or stub note into a full draft, fleshing bullet points into prose, or drafting longer-form writing grounded in vault context and the user's voice.

Use when turning a planning note into a build spec, preparing a note for handoff to a coding-agent build, or creating a spec and tracker from a plan.

Use when running a formal research project on typed research records — creating a project, importing sources, capturing evidence with locators, building supported claims, auditing provenance, and generating an evidence-backed outline via the research_* tools.

Use when distilling or saving a Claude session into the Obsidian vault, capturing what was decided or learned in a session, or turning session memory into a persistent, linked knowledge note.

Use when comparing notes that are typed or tagged as sources or papers, building an evidence or comparison table across them. Operates on existing plain source notes; typed provenance-record workflows are outside the portable CLI scope.

Use when adding a TL;DR or summary to a long note, condensing a note, or surfacing and linking the key concepts inside a note.

Use when collecting open tasks or todos scattered across notes into one consolidated action list — unchecked checkboxes and #task/#todo-tagged items across the whole vault, regardless of when the notes were written. For a review bounded to a recent time window, use daily-rollup.

Use when driving a spec-based build against the vault and updating its tracker note, or reporting build progress honestly to a tracker as tasks complete.

Use when another obsidian-agent skill declares it as a required sub-skill — the honesty rules for citing, linking, tagging, and writing in a vault. A shared discipline other skills invoke, never a response to a user request on its own.

Use when a vault result has recurring value and the user wants it repeated on a schedule. Registers a scheduled cloud agent in Claude Code; in Companion it can only fire a routine the user already created.

Use when answering "what do I know about X" from the whole vault — a grounded, cited topic synthesis across all notes regardless of type, with contradictions and gaps named.

Use when a note's body mentions other notes by title without linking them, when applying or repairing wikilinks, or when listing orphan notes. This is the skill that edits notes to add links. For conceptual discovery instead, use connection-finder.