name: horadus-cli
description: Use when operating the Horadus repo through task, sprint, triage, eval, or branch-start workflows; prefer the horadus CLI over ad hoc markdown parsing or one-off shell helpers when an equivalent command exists.
Horadus CLI
Use this skill for repo workflow operations in this project. For command details,
read references/commands.md; for lifecycle policy and merge/review semantics,
read AGENTS.md.
Implementation note:
- Canonical CLI ownership lives under
tools/horadus/python/horadus_cli/. - The installed
horadusentrypoint points directly at the tooling package. - App-backed commands cross
tools/horadus/python/horadus_app_cli_runtime.py; the tooling package should not import business-app modules directly.
Default behavior
- Prefer
horadusover directrg/awk/markdown scraping when the CLI covers the workflow. - Prefer
--format jsonfor agent use. - Prefer
--dry-runbefore any branch-creating command. - Use raw
git/ghcommands only when the Horadus CLI does not expose the needed workflow step yet, or when the CLI explicitly tells you a manual recovery step is required. - If a forced fallback is still required after those recovery attempts,
record it with
horadus tasks record-friction; do not log routine success cases or expected empty results. - Fall back to repo files or legacy scripts only when the CLI does not expose the needed surface.
Command Surfaces
Root horadus command groups to check before falling back:
uv run --no-sync horadus trends statusuv run --no-sync horadus dashboard exportuv run --no-sync horadus eval benchmarkuv run --no-sync horadus pipeline dry-runuv run --no-sync horadus agent smokeuv run --no-sync horadus doctoruv run --no-sync horadus tasks list-activeuv run --no-sync horadus triage collect
Core task lifecycle commands:
- Start preflight:
uv run --no-sync horadus tasks preflight - Dirty-main watchdog:
uv run --no-sync horadus tasks assert-safe-worktree - Canonical autonomous start:
uv run --no-sync horadus tasks safe-start TASK-XXX --name short-name - Context pack:
uv run --no-sync horadus tasks context-pack TASK-XXX --mode implement --format json - Close task ledgers in-branch:
uv run --no-sync horadus tasks close-ledgers TASK-XXX - Fast iteration gate:
make agent-check - Pre-push local review:
uv run --no-sync horadus tasks local-review --format json - Canonical local gate:
uv run --no-sync horadus tasks local-gate --full - Lifecycle verifier:
uv run --no-sync horadus tasks lifecycle TASK-XXX --strict - Finish:
uv run --no-sync horadus tasks finish TASK-XXX
Common follow-up and operations commands:
- Capture follow-up intake:
uv run --no-sync horadus tasks intake add --title "..." --note "..." - Promote intake only during deliberate grooming:
uv run --no-sync horadus tasks intake promote INTAKE-XXXX --priority ... --estimate ... --acceptance "..." - Automation lock health:
uv run --no-sync horadus tasks automation-lock check --automation-id <id> - Code-health eval:
uv run --no-sync horadus eval code-health
Eval command reminders:
- Tier 1-only benchmark diagnostics:
uv run --no-sync horadus eval benchmark --tier-scope tier1 - Tier 2-only benchmark diagnostics:
uv run --no-sync horadus eval benchmark --tier-scope tier2 - Behavior evals:
uv run --no-sync horadus eval behavior - Taxonomy validation:
uv run --no-sync horadus eval validate-taxonomy - Regression intake:
uv run --no-sync horadus eval regression-intake - Vector benchmark:
uv run --no-sync horadus eval vector-benchmark - Embedding lineage:
uv run --no-sync horadus eval embedding-lineage - Source freshness:
uv run --no-sync horadus eval source-freshness
When to read more
- For command examples and output expectations, read
references/commands.mdordocs/AGENT_RUNBOOK.md. - For canonical workflow policy and completion rules, read
AGENTS.md. - For high-risk cross-surface tasks (for example migrations, shared workflow
tooling or config, shared math, or multi-surface mutation work), front-load
adversarial review before the first push instead of discovering the whole bug
set inside
horadus tasks finish. - If
horadus tasks context-pack TASK-XXXrecommends pre-push local review, follow that guidance. The default/env provider chain already falls through missing provider CLIs on PATH in repo order. If the first local-review run hits a provider-specific timeout, auth/config failure, or unreadable output and you still want local automation, rerun with--allow-provider-fallback; if the local-review path still remains unusable, request manual review early rather than waiting for the finish loop. - Batch related fixes with updated tests before re-requesting review on a high-risk task; do not turn the same open bucket into a single-commit re-review loop.
- Provider selection for local review is:
--provideroverride first, thenHORADUS_LOCAL_REVIEW_PROVIDERfrom optional local-only.env.harness, then the repo defaultclaude.