Skip to content

Releases: machine-machine/herdr-factory-loop-skill

v2.8.0 — Prime Agent worker

Choose a tag to compare

@mariuszkreft mariuszkreft released this 11 Aug 10:15
6b4ee62

Adds Prime Agent (Prime Intellect's open-source harness) to the worker selection.

Added

  • prime worker agent across the factory: m2herd-up dispatch --agent prime (pane + --headless), settings routing, SKILL §9.1 worker table.
  • Headless: prime-agent -p --session-dir .m2herd/dispatch/<slice>.prime-session (plain text mode, log-tail salvage; --model only when explicitly set). Resume: watch respawns with -p -c --session-dir <same>.
  • No auto-approve flag needed — prime has no per-tool approval gate (verified against binary v0.7.0).
  • goal_support.md: prime row — launch-time --goal + --autonomous gates; fleet-loop re-nudges on idle.

Notes

  • No herdr integration for prime yet — pane lifecycle is heuristic; prefer --headless.
  • Requires Node ≥ 20.6 (curl -fsSL https://app.primeintellect.ai/prime-agent/install.sh | sh).
  • Token/cost parsing deferred until prime's usage schema is verified — "no cost recorded" stays distinct from "free".
  • This release also ships the previously untagged 2.7.2 changes (pi worker, .m2herd/graph.json declared topology — see CHANGELOG).

What's Changed

  • fix(m2herd-up): worker panes stray to new workspaces on bind-mounted homes by @mariuszkreft in #16
  • fix(m2herd-up): stop reporting a pointer that never arrived as a successful dispatch by @mariuszkreft in #17
  • feat: add Prime Agent (prime) to the worker selection (2.8.0) by @mariuszkreft in #18

Full Changelog: v2.7.1...v2.8.0

v2.7.1

Choose a tag to compare

@github-actions github-actions released this 29 Jul 22:36

Full Changelog: v2.7.0...v2.7.1

v2.7.0

Choose a tag to compare

@github-actions github-actions released this 09 Jul 14:54

What's Changed

  • feat: compound-engineering pass — review stage, run reports, lesson promotion (v1.2.0) by @mariuszkreft in #1
  • feat: SDD factory loop — spec-kit × herdr + orchestrator onboarding (v1.3.0) by @mariuszkreft in #2
  • feat: cursor orchestrator/worker + ICM-steered herd loop (v1.5.0) by @mariuszkreft in #3
  • fix(herd-loop): make live worker dispatch work end-to-end by @mariuszkreft in #4
  • fix(herd-loop): self-healing prompt resend + dead-worker detection by @mariuszkreft in #5
  • feat(meta): meta-orchestrator tier — fleet-loop.sh + fleet-control/ (orchestrator of orchestrators) by @mariuszkreft in #6
  • fix(dispatch): settle before Enter so prompts don't sit unsubmitted by @mariuszkreft in #7
  • feat(hooks): default-on dispatch nudge for Claude Code + Hermes (v1.7.0) by @mariuszkreft in #10
  • feat(context): budget-aware task decomposer + context-offloading Hermes hooks by @mariuszkreft in #9
  • feat(m2herd): the Fable main-orchestrator context fabric (v2.0.0) by @mariuszkreft in #11
  • feat(m2herd): headless worker tier — cheap hands, Fable judgment (v2.2.0) by @mariuszkreft in #12
  • feat(m2herd): auto-kick — the fabric wakes its own orchestrator (v2.3.0) by @mariuszkreft in #13
  • feat(m2herd): tier-3 Go TUI — bubbletea dashboard, fancy on any terminal (v2.4.0) by @mariuszkreft in #14
  • feat: continual harness factory evolver (MVP) by @mariuszkreft in #15

New Contributors

Full Changelog: https://github.com/machine-machine/herdr-factory-loop-skill/commits/v2.7.0