Repository navigation
Releases: avalonreset/legends-dataforseo
Release list
legends-dataforseo v0.1.6
legends-dataforseo v0.1.6
- Raise minimum supported Python runtime to Python 3.11+ (
requires-python = ">=3.11"). - Harden GitHub Actions CI workflows with 40-character commit SHA action pins, explicit
contents: readpermissions, andpip>=26.2.2before package installation.
legends-dataforseo v0.1.5
legends-dataforseo v0.1.5
Highlights
- Google Business Profile (
my_business_info) support:- Added
business()Python helper andlegends-dataforseo businessCLI subcommand for live Google Business Profile lookups bycid:,place_id:, or raw keyword (--location-code,--location-name,--location-coordinate,--tag). - Added
business($0.002baseline live estimate),business-task-post, andbusiness-task-getroutes toroutes.json. - Enabled
wait_task()andlegends-dataforseo waitpolling on/business_data/google/my_business_info/task_get/<id>. - Updated
docs/INTEGRATIONS.md,docs/API.md,docs/CLI.md, anddocs/INSTALL.md.
- Added
Verification
- Commit:
1e602c00f6cd654329d696e2731548948776479b - Test suite:
269passed - Public source & archive audit (
scripts/audit_public.py): passed - Clean venv wheel smoke test (
scripts/smoke_wheel.py): passed
legends-dataforseo v0.1.4
The repository, Python distribution and CTO Legends catalog identity are now legends-dataforseo.
- Canonical repository: https://github.com/avalonreset/legends-dataforseo
- CLI and Python import remain
legends-dataforseoandlegends_dataforseo. - Existing documentation caches and cost ledgers are reused in place.
- When upgrading a standalone virtual environment, uninstall the former
legends-dataforseo-kitdistribution before installing this wheel, since both provide the same import and executable. - CTO Legends retains the old command identifier as a compatibility alias; manager 0.2.2 recognizes retained installations without moving them.
Validation: offline test suite, clean-wheel install and CLI/evidence smoke, source/archive audit, and Linux/Windows/macOS CI.
legends-dataforseo 0.1.3: reusable evidence workflows
Save complete DataForSEO responses, find them later and check their integrity and reuse eligibility without another provider call.
- Adds
evidence inventory,find,viewandreuseto the primary CLI. Existing export/verify entry points remain supported. - Binds the intake note as well as raw response for new packages. Older v1 packages remain readable with explicit weaker note-integrity coverage.
- Preserves original response bytes and full output by default. Focused display is optional.
- Rejects ambiguous JSON and linked evidence paths. Pending, failed, stale and wrong-workspace evidence cannot silently substitute for completed current research.
- Documents portable client workspaces and the reviewed Empire intake workflow. No bypass, savings percentage or automatic strategy claims.
Validation: 263 tests passed on Windows, one symlink-privilege skip covered by independent Linux checks; wheel/sdist, public audit, Twine and isolated installed-wheel smoke passed. An independent 18-case cross-module matrix and 19-call Empire capture/knowledge rehearsal passed without provider requests.
Synthetic measurement: a 200-package inventory opened zero raw responses; an 8,403,634-byte response produced a 659-byte explicitly selected view while preserving the full original. This is local evidence handling, not a provider throughput guarantee. Views still parse a selected response in memory.
v0.1.2 version join
Fix: package version and routes registry version join 0.1.2. No behavior changes.
v0.1.1 docs baseline
Docs-only patch: pre-reset version references reworded to the 0.1.x router-native baseline; Legends Obsidian references renamed to legends-empire. No code changes.
legends-dataforseo-kit v0.1.0
Router-native generation reset (reset generation 2026-09-25).
- Vendored router skill: skills/cto-legends/SKILL.md, byte-exact vs cto-legends at router pin 6975dcbf073e5eec7a30ce6a04a6975238c61158.
- Per-module skill registration (skills/legends-dataforseo-kit/) removed.
- Version reset to 0.1.0; git history preserved; prior releases and tags wiped.