PanzerLlama/youtrack-aitrack v0.4.0b0

v0.4.0b0 — partial re-runs, tokens + cost, repo visibility

on GitHub

Iterating on a single report is now cheap and visible: re-run one action
instead of the whole workflow, see what it cost, and see which repo and
branch the run actually inspected.

Added

• yta run --action ID (repeatable) re-runs only the named action(s).
  The selection is expanded to its transitive depends_on chain, so
  --action qa_plan regenerates pages_changed first and the downstream
  report sees fresh upstream output. Workflow on_success / on_failure
  hooks are skipped on a partial run. Requires --workflow; an unknown id
  fails before any I/O and lists the valid ids. Pure helper:
  select_actions in domain/workflow.py.
• Per-action tokens and USD cost. The claude_code_cli backend now
  spawns claude -p --output-format json and reads the result envelope.
  AgentResult gains optional input_tokens, output_tokens, cost_usd;
  ai_report stores them in its output; the yta run summary gains
  TOKENS (in/out, input includes prompt-cache writes and reads) and
  COST columns plus a per-workflow total on the === line. Blank for
  --stub-llm and for backends that do not report usage.
• yta run prints the repo and branch it operates on: repo: <path> (cwd; pass --repo-dir to override) before dispatch, then branch: <name> @ <sha> or branch: none (pattern '…'); … Wrong repo? pass --repo-dir
  before the summary table. RunReport persists repo_dir, branch,
  commit_sha so a stored run is self-describing.

Fixed

• --dry-run notes no longer claim side effects. git_branch,
  write_file and bd_issue report [dry-run] would create … / would write … / would create beads issue. The dry-run branch adapter reads
  the real repo state, so "would reuse " and the dirty-tree refusal
  are accurate while create/switch stay no-ops.
• yta run / yta poll show a one-line error instead of a traceback
  when a workflow YAML fails to parse (for example an action type unknown to
  an older installed version).
• The Claude CLI runner raises AgentRunnerError with the envelope's
  subtype when the CLI reports is_error on a zero exit code.

Internal

• Dry-run adapters moved from runtime/factory.py to runtime/dry_run.py;
  yta run summary rendering moved to cli/run_summary.py; the JSON
  envelope contract lives in adapters/cli/envelope.py.

Install

uv tool install --from git+https://github.com/PanzerLlama/youtrack-aitrack@v0.4.0b0 youtrack-aitrack