Skip to content

Autonomy hardening + generative goal pursuit (v0.14.0 → v0.16.0) - #17

Merged
scr1p7k177y merged 4 commits into
mainfrom
feat/openai-compatible-adapter
Jul 18, 2026
Merged

scr1p7k177y merged 4 commits into
mainfrom
feat/openai-compatible-adapter

Conversation

@scr1p7k177y

Copy link
Copy Markdown
Member

Brings main up from v0.14.0 to v0.16.0. Five stacked commits, all previously reviewed; 232 tests + eslint green locally.

What's included

  • v0.14.1 — offline demo + sample-usage docs (examples/demo.mjs).
  • v0.14.2 — reach any OpenAI-compatible endpoint via base_url.
  • v0.15.0 — autonomy hardening. Wired the documented-but-dead governance gates (source tasks default to needs-approval, opt-in executors.require_approval, ceo_workflow budget-gated); crash/stale-claim recovery; stop re-dispatching a completed task on a post-dispatch throw; requeue on budget-approve; windowed per-session cap; $0 non-LLM spend; base_url TLS guard; safe task-id/workflow-name; agent roster 4→7; untracked committed db/ + stale temp/.
  • v0.16.0 — pursue goals autonomously. lib/strategist.js turns strategy/goals.md into the next tasks when the queue is idle (default OFF, dry-run/needs-approval by default), gated by the existing pipeline and self-bounded by per-day task+plan caps and a re-plan cooldown. comms/STOP kill switch (lib/killswitch.js) halts all dispatch. Plus the re-review guards (per-step STOP in workflows, state-safe approval resolution).

Review + tests

  • Each feature got a Fable /code-review; every finding fixed. v0.16.0 got a second re-review (APPROVE-WITH-NITS) whose 2 guards are also fixed.
  • 232 tests (unit + integration + stdio e2e, offline mock provider) + eslint green.

Notes

  • Clean fast-forward of main (feat is a direct descendant).
  • The docs/sample-usage-and-offline-demo branch is already folded in here and can be deleted after merge.
  • Everything risky stays default-OFF and approval-first.

Please review — not requesting auto-merge.

https://claude.ai/code/session_01CKgPgAMauJxTHj7AM9ZxjW

…counting (v0.15.0)

Wire the documented-but-dead governance gates (source tasks default to needs-approval,
opt-in executors.require_approval, ceo_workflow budget-gated); recover crashed/stale
in-progress claims; stop re-dispatching a completed task on a post-dispatch throw;
requeue on budget-approve; window the per-session cap; $0 non-LLM spend; base_url TLS
guard; safe task-id/workflow-name; fix agent roster; untrack committed db/ and temp/.
README version badge + test count (213), tools/index.json and settings.yaml
version fields, and correct the ceo_recall semantic wording to FTS.
New lib/strategist.js turns strategy/goals.md into the next tasks when the queue
is idle (default OFF, dry-run/needs_approval by default), gated by the existing
budget/approval/veto pipeline and self-bounded by per-day task+plan caps and a
re-plan cooldown. Add a comms/STOP kill switch (lib/killswitch.js) honored by the
heartbeat, runner, and workflow. Rejecting a gated task is now terminal; pending
dry-run tasks no longer consume real-work cycle slots. 228 tests + lint green.
…solution

Follow-up to the v0.16.0 re-review. Kill switch now re-checked per workflow step
(a mid-run STOP halts remaining steps). Resolving an approval no longer flips a
task that changed state: reject blocks only a still-pending task; budget-approve
requeues only a still-blocked one. +4 tests (slot-starvation, mid-workflow halt,
reject-terminal + done-not-flipped). 232 tests + lint green.
@scr1p7k177y
scr1p7k177y merged commit 511e02b into main Jul 18, 2026
2 checks passed
scr1p7k177y added a commit that referenced this pull request Oct 4, 2026
Autonomy hardening + generative goal pursuit (v0.14.0 → v0.16.0)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant