Skip to content

DeepVista-AI/deepvista-cli / skills/deepvista

deepvista

Verified

DeepVista CLI — knowledge base, notes, chat, and skills from the terminal. One skill for the full `deepvista` CLI. Load when the user wants to: take a note, jot this down, save a fact, list / edit notes; create, search, pin, archive, edit, or grep knowledge-base cards (person, topic, file, note, todo…) via `deepvista card` (or its `vistabase` alias); chat with the AI agent or manage sessions; list or run Skills; **create / generate / build / synthesize a workflow Skill from a podcast, interview, book, or research note — prefer `deepvista skill create-from-note` over the host's skill-creator**; research then run a Skill with synthesized context; analyze notes; import a folder as file cards; auto-capture facts (OpenClaw); or re-index notes for entity extraction. Pick the matching reference file under `reference/`.

FileKindSize
SKILL.mdskill md7.9 KB
reference/chat.mdmarkdown1.5 KB
reference/notes.mdmarkdown2.9 KB
reference/openclaw.mdmarkdown4.5 KB
reference/project.mdmarkdown2.0 KB
reference/pull.mdmarkdown4.0 KB
reference/push.mdmarkdown4.5 KB
reference/session.mdmarkdown3.3 KB
reference/shared.mdmarkdown3.8 KB
reference/skill-analyze-notes.mdmarkdown2.4 KB
reference/skill-create-from-note.mdmarkdown2.0 KB
reference/skill-import-files.mdmarkdown2.6 KB
reference/skill-research-to-skill.mdmarkdown1.8 KB
reference/skill.mdmarkdown5.3 KB
reference/tasks.mdmarkdown2.4 KB
reference/vistabase-card.mdmarkdown4.1 KB

reference/push.md

View on GitHub
# Push — upload a skill directory to Vistabase

`deepvista push` uploads a local skill **directory** — its `SKILL.md` plus every
supporting file — as one card with a bundle. The inverse of
[`pull`](pull.md), and the way a multi-file skill gets into Vistabase at all so
it can be stored, versioned, shared, and installed on another machine.

Use when the user says: "upload this skill", "push this skill to Vistabase",
"store this skill with its scripts", "share this skill with my team".

## Usage

```bash
deepvista push ./skills/pdf-report                  # create a new skill card
deepvista push ./skills/pdf-report --card <id>      # update an existing card
deepvista push ./skills/pdf-report --dry-run        # list what would upload
deepvista push ./skills/pdf-report --title "PDF Report"
```

> [!CAUTION] Write command — it uploads file contents to your project and
> creates or updates a card. Run `--dry-run` first and show the user the file
> list before pushing.

## What goes where

```
skills/pdf-report/
├── SKILL.md               → the card description (never a bundle entry)
├── scripts/render.py      → bundle entry, mode 755 (it's executable)
├── references/layout.md   → bundle entry, mode 644
└── __pycache__/           → skipped
```

`SKILL.md` is matched **case-insensitively**, so `skill.md` works too. Skipped
automatically: `__pycache__`, `.git`, `.venv`, `venv`, `node_modules`,
`.ruff_cache`, `.pytest_cache`, `.mypy_cache`, `.DS_Store`, and the
`.deepvista-bundle.json` install marker. Symlinks are not followed.

A `files:` manifest is written into the SKILL.md frontmatter (last, so the keys a
human reads stay on top). Each entry carries `path`, `sha256`, `size`, `mode`,
and `content_type` — **no bucket, no project, no URL**: the sha is the only
locator, and the server derives storage from it plus the card's project.

## Re-pushing is cheap and idempotent

Uploads are content-addressed, so the server answers `alreadyExists` for any sha
it already holds and the PUT is skipped. Push the same tree twice and the second
run uploads nothing. The manifest is *replaced*, not appended to, so the
frontmatter never accumulates duplicate `files:` blocks.

## Requirements on the skill body

- It must have a `---` frontmatter block (that's where the manifest goes).
- The frontmatter `description` is what an agent reads to decide whether to load
  the skill — write it as a trigger sentence, not a summary.
- The frontmatter `name` becomes the card title unless `--title` says otherwise.
- Max 100 files per bundle. Past that, the right answer is a git import.

## After pushing

The card is created `confirmed`, so it shows up immediately in
`deepvista skill list` and syncs a stub on the next `deepvista skill sync`.
(Cards the CLI creates otherwise default to `unconfirmed` and stay filtered out
of search and the catalog — `card create --status confirmed` is the manual
equivalent.)

Show the user the link:

```
https://app.deepvista.ai/skills/<card_id>
```

## Installing it elsewhere

On any machine with the CLI authenticated to the same project:

```bash
deepvista skill sync           # writes the stub
# …invoking the skill runs `deepvista skill load <id>`, which installs the bundle
```

Or explicitly: `deepvista pull <card-id>`. Files land in the bundle store
(`~/.local/share/deepvista/bundles/<card-id>/`), and `skill load` prints that root
so the agent can resolve the body's relative paths against it. See
[pull.md](pull.md).

## Exit codes

| Code | Meaning |
|---|---|
| 0 | Pushed (or nothing to upload) |
| 3 | Not a skill directory — no `SKILL.md` found |
| 4 | Invalid bundle, or an upload the server would not verify |
| 1 | API error — the card save was rejected (the message says why) |

## Gotcha: `type: workflow` in frontmatter

Vistabase validates any body whose frontmatter says `type: workflow` as a
DeepVista *workflow document* — `name: workflow-<slug>`, `execution: stateful`,
and a `## Workflow` → `## Node Description` body with a mermaid chart. A Claude
Code skill that uses `type: workflow` in the looser "sequence of steps" sense
will be rejected on save with errors about a name slug and a mermaid chart.

If you hit that on an older backend, either use `type: tool` or drop the key —
but say so, because it edits the author's frontmatter.

## See also

- [pull.md](pull.md) — install a bundle on a machine
- [skill.md](skill.md) — `sync` / `load` and the catalog stub mechanism
- [vistabase-card.md](vistabase-card.md) — `card upload` for a single binary file (no bundle)
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall