Understand a Bubble application through its export. SpecItOut turns a private export into searchable entities, relationships and source evidence so your coding agent can investigate dependencies and explain what the exported app contains.
Use it when you are taking over an unfamiliar app, tracing a workflow, assessing a proposed change, or preparing a migration. The CLI retrieves evidence; your agent’s LLM writes the answer.
Give your coding agent this repository URL and paste:
Install SpecItOut and its analysis skill from this repository. Set up a private workspace for my Bubble application, then ask me for the local export path. Process the export with progress updates and verify readiness before answering questions about it.
The agent follows the setup guide, installs the CLI and bundled skill from the same Git revision, and creates a separate private folder for your app. Git is not required in that folder. One CLI installation can serve several workspaces.
Version: 0.2.2. Install from an explicit repository commit using the setup guide; a package-index release is not assumed. The initial native-runtime target is macOS or Linux as a non-root user, with Python 3.11+, Postgres, pgvector and sentence-transformers. Native Windows, Docker and hosted databases are outside this onboarding path. See the setup guide for prerequisites and validation limits.
The Python CLI parses the export, stores a versioned application graph in local
Postgres, builds search documents, and computes local embeddings with
sentence-transformers and BAAI/bge-small-en-v1.5. The diagram shows a CPU
onboarding profile; the provider otherwise selects its device automatically.
An initial model download is separate from work on each export. Processing time
is unbenchmarked: use measured elapsed time and counts, not a promised duration.
Processing stages, storage and limitations →
This is retrieval-augmented generation (RAG): the CLI retrieves evidence, then the agent generates an answer grounded in that evidence.
Exact identifiers, lexical matches and vector similarity help find an anchor. Graph traversal then exposes connected entities. A bounded bundle gives the agent relevant evidence and explicit limits without loading the entire export into the conversation. The CLI computes locally; a hosted agent’s LLM can receive retrieved application evidence under that agent provider’s data handling.
Try questions such as:
- “Where is this field referenced, and which connected actions may write it?”
- “What uses this workflow, and what does it depend on?”
- “Which pages are connected to this reusable element?”
- “What paths connect these two entities?”
- “Collect the evidence relevant to moving this workflow out of Bubble.”
These questions map to search, entity lookup, usage, paths and context bundles. The answer depends on what the parser extracted and what the selected version contains. Business intent, runtime execution and behavior missing from the export require additional evidence.
An answer should look like this synthetic illustration, not a claim about a real application or a literal CLI response:
Observed:
approve_orderreferencesOrder.statusthrough actionset_status. [1]Interpretation: This may be part of order approval. The export alone does not establish the business policy or prove that this path runs in production.
Source: [1] export version
demo-v1, actionset_status, source path$.workflows.approve_order.actions.set_status.Limit: This conclusion covers the retrieved references; check bundle truncation and unresolved references before claiming completeness.
The source reference lets a reviewer return to the stable export version. The separation between observation and interpretation makes the conclusion reviewable.
my-bubble-app/ private folder; no Git required
specitout.toml app identity
exports/ stable, versioned copies of your exports
.specitout/ database files, model cache and runtime state
reports/ optional requested analysis outputs
Importing preserves your original file. Uninstalling the CLI preserves private workspaces. Stopping the database, removing the skill and deleting a workspace are separate actions. Folder separation is organizational: the native localhost Postgres runtime uses trust authentication and is not a security boundary against other local users or processes.
- Usage, command examples and recovery
- Uninstall and data lifecycle
- Agent installation guide
- Versions, changelog and release process · Changelog
- Validation performed and remaining acceptance checks
- Public data policy
The CLI analyzes exports; it does not edit or execute your live Bubble app. This repository ships no Bubble exports or client-derived artifacts. A private real-export acceptance run has completed locally; the validation notes describe its coverage, verified corrections and remaining limits. One successful export does not establish support for every Bubble export.