Skip to content
P‑AIProject‑AI
Systems · TAAR

TAAR

Thirsty's Active Agent Runner (TAAR) is documented as a local-first governed taskforce with evidence bundles and report-first execution boundaries.

What TAAR is

TAAR is presented as a local-first governed taskforce that separates reader agents from writer agents, enforces registry-declared capabilities, and writes append-only audit records.

Governed taskforce model

  • Reader-agent and writer-agent separation
  • Registry and capability enforcement before command execution
  • Evidence bundles and append-only audit behavior
  • GitHub Workflow Guardian for workflow risk checks
  • Classification model from OPEN to BLACK

Current package version

taar-agent-taskforce 0.1.0

Source: pyproject metadata at pinned commit.

Open metadata source

What TAAR Does Not Do

Canonical README boundary: the first swarm is report-only. It does not autonomously apply patches, merge, deploy, publish, delete, or mutate repository state.

Any stronger operational claim remains Unverified unless a newer canonical source supersedes this boundary.

Status, limitations, and proof trail

  • Documented Governance model and workflow guardian surfaces.
  • Verified Package metadata version and pinned source reference.
  • Unverified Autonomous mutation/deployment capabilities (explicitly out of scope in current docs).
README proof sourcePinned repository revisionRepository