Use Landing locally¶
Run a task against your own checkout with Python 3.12 or later, uv, and a POSIX host. Install Landing as an isolated tool, then prepare your project dependencies and checks:
uv tool install landing
export LANDING_MODEL="openai:gpt-4.1"
export LANDING_API_KEY="your-provider-api-key"
This installs the latest release. Update it with uv tool upgrade landing. For a development build from main, use uv tool install --force git+https://github.com/bubbuild/landing.git; its version identifies the source revision.
Run commands from your project checkout. If landing is not on your PATH after installation, run uv tool update-shell and restart your shell. For another provider or endpoint, see Configuration.
Explain a failure¶
Use an existing UTF-8 failure log and select the project workspace:
landing explain "Explain the failure and suggest the next useful check." --workspace ../project --input check.log
Read whether the answer establishes a cause or proposes a hypothesis. Use triage when you need reproduction evidence and acceptance criteria for a work item. Without a configured platform capability, results stay in the terminal.
Delegate a fix¶
Prepare a disposable checkout or branch you can inspect. State the expected user behavior and supply the check that establishes it:
landing fix "Fix the missing-name behavior: print usage to stderr and exit 2. Preserve valid greetings." --workspace ../candidate --input check.log --check "python acceptance.py" --json --output fix.json
Replace the instruction and check with your actual problem. Fix changes the workspace, then Landing runs the required check. A successful action returns status: completed; failed execution or validation preserves partial changes and any explanation. Inspect the diff and the recorded check before accepting it. Ordinary local execution does not automatically commit or publish.
Delegate a dependency update¶
Name the dependency, target version, directory and applicable project policy. Prepare its update tools and an isolated checkout:
landing bump "Update example-lib to exactly 1.2.3 using project tools and .github/dependabot.yml. This is local work without CI; repair demonstrated compatibility failures." --workspace ../candidate --check "make acceptance" --json --output bump.json
Replace the target, configuration location and check with your request. An exact target can change the manifest version while preserving its pinning style; a lockfile refresh stays within existing constraints. The agent follows project policy, updates dependency files and runs local checks. Without CI, demonstrated check failures can guide repair in the same invocation. Preparation-only work leaves the candidate diff and results. Ordinary local execution does not commit or publish.
Required final checks run after agent work; failure preserves edits and the result without restarting repair. Inspect the resolved version, diff and checks before accepting the candidate. For GitHub updates, compatibility repair waits for current candidate CI failure.
Review the candidate¶
landing review "Review the candidate against the issue's acceptance criteria." --workspace ../candidate --check "python acceptance.py" --json --output review.json
Landing runs the check before review. A failed check forces block; missing evidence can produce inconclusive. The command exits zero only for a completed review with allow. Review guidance asks the agent to leave the candidate unchanged; configured tool lists filter capabilities but are not a sandbox.
Inspect and continue¶
landing action list
landing action view act_example --json
landing action logs act_example
Replace act_example with an ID from the list. Use the same database as the task; --db or LANDING_DB selects it. Workspace edits remain in the selected checkout. See CLI reference for piping input, cancellation, retries, and exit codes.