capivaOS 1.5.0

capivaOS 1.5.0

The autonomy-workflow release, plus the distribution fixes from a live install debugging session.

Added

• Two-sprint cycle (RFN-013) — /capiva:refine (grill-sprint) → /capiva:auto (execution-sprint) documented as one flow, with a board-state handoff and a single interactive pre-flight confirmation on /capiva:auto. Scheduled runs skip it via the unattended token / Auto Preflight: off. ADR-0014 4th amendment.
• Docs-generation step (RFN-012) — the execution workflow auto-generates a project-facing feature doc (docs/features/<TASK-ID>.md) per task; non-blocking, with a self-tested validator.
• Refine batch-cadence + Step-3 clarity (RFN-011).

Changed

• Invocation-reference convention (DIST-005, ADR-0019) — skills invoke under their unqualified name (/sprint); the /capiva: prefix appears only when a name collides with a built-in (/capiva:init). Engine deny messages + skill descriptions updated; harness_lint relaxed to validate resolution instead of mandating the prefix.
• Invocation-syntax docs (DIST-003) — README + troubleshooting explain the above (fixes the first-run "/capiva:sprint → no matching command" confusion in the VS Code extension).

Fixed

• Update-ritual command (DIST-004) — /capiva:update Step 3 corrected to claude plugin update capiva@capiva.

Full detail: CHANGELOG. Update with claude plugin update capiva@capiva.

v1.4.0 — RFN epic (batch-refine / cluster-HITL)

Cluster grilling up front, run execution unattended, review once (RFN-001..009). Everything is opt-in and default-off; attended /capiva:sprint is unchanged. The batch-refine workflow (/capiva:refine + REFINING + context-answerer) ships EXPERIMENTAL — validated by the RFN-002 eval gate + a tabletop dry-run, not yet live-run. Full details in CHANGELOG [1.4.0].

capivaOS 1.3.0 — the proof-of-life release

What's Changed

• chore(hygiene): public repo carries no private history — lint-enforced by @iB2 in #42
• feat(guard): PR-A — guard liveness + write-surface + state integrity (PRD-001/002/003) by @iB2 in #43
• feat: PR-D — post-epic audit follow-ups (PRD-007/008/009) by @iB2 in #46
• feat: consolidation — land PRD stack B+C+D on main (recovers #44/#45/#46 cascade) by @iB2 in #47

Full Changelog: v1.2.1...v1.3.0

capivaOS 1.2.1 — the grilled remainder

What's Changed

• fix(sweep): grilled leftovers — 8 survivors, 2 rejections, 1 refutation #AUD-018 by @iB2 in #39
• fix(engine): private board IDs out of shipped content + lint check 15 #AUD-020 by @iB2 in #40
• chore(release): 1.2.1 — the grilled remainder by @iB2 in #41

Full Changelog: v1.2.0...v1.2.1

capivaOS 1.2.0 — the honesty release

What's Changed

• fix(manifest): remove keys the CLI validator rejects + real-install CI gate #AUD-001 by @iB2 in #21
• chore: land the AUD-002..004 stack into main by @iB2 in #25
• feat(claims): ENFORCED_SURFACES — code and claims mechanically locked #AUD-011 by @iB2 in #32
• docs: comparison table — source-verified, dated, caveats included #AUD-012 by @iB2 in #33
• docs(adr): ADR-0015 — transition validation design, implementation deferred #AUD-013 by @iB2 in #34
• feat(hooks): compaction counter — the 2-compaction rule becomes evaluable #AUD-014 by @iB2 in #35
• feat(lint): --check-blueprint — the authoring contract gets tooling #AUD-015 by @iB2 in #36
• docs: CONTRIBUTING — state what single-owner CODEOWNERS implies #AUD-016 by @iB2 in #37
• chore(release): 1.2.0 — the honesty release #AUD-017 by @iB2 in #38

Full Changelog: v1.1.0...v1.2.0

capivaOS 1.1.0 — the autonomy release

capivaOS can now work a pre-approved backlog unattended — under a contract, not a vibe.

Update

/capiva:update        (or: tell Claude "update the harness")

No project-file migration required (optional new config fields: Phase Isolation, Dual Review, Auto Task Cap — all default to today's behavior).

Auto mode (ADR-0014)

/capiva:auto

• Works only tasks whose specs YOU approved (or fast-lane qualifiers) — it never grills, never improvises scope
• Gates route through your handwritten .board/approval-policy.md + an independent read-only judge; exceptions queue in .board/approvals.md with one-paragraph summaries
• The never-list (no machine may clear, not configurable): merge decisions, P0/P1 gates, approval of human-less specs, anything your policy doesn't explicitly cover
• The policy file is hook-protected from agent edits — self-licensing is mechanically impossible
• Hard budgets always (default 3 tasks/run); parks cleanly at phase boundaries with a handover doc; morning report leads with why it stopped
• Still zero network, zero deps, zero telemetry — escalation is a file, notification is your own hook if you want one

Also in 1.1.0

• Phase Isolation (opt-in for attended runs): each phase executes in a fresh context fed only by the on-disk artifacts — long sessions stop degrading
• Loop state survives compaction; full board snapshots to .state/ on every save
• Dependency-aware board: triage skips blocked tasks deterministically; cycles are lint-caught defects
• /capiva:init now checks GitHub branch protection and offers to configure it — the merge gate is only real if the forge enforces it
• Optional dual independent review; reviewer disagreement forces escalation

Attended mode is byte-for-byte unchanged. Full details: CHANGELOG.md · MIT

capivaOS 1.0.0 — spec-driven development harness, now a Claude Code plugin

First plugin release. capivaOS is a state-machine enforced, spec-first, test-enforced development pipeline for Claude Code — free, by Capiva.

Install

/plugin marketplace add iB2/capivaOS
/plugin install capiva@capiva

Then run /capiva:init inside your project. Update any time with /capiva:update.

Highlights

• Engine/state split: skills, hooks, agents, rules, and blueprints live in the plugin and update centrally; /capiva:init scaffolds your project's board + docs and stamps a schema version
• Six-phase pipeline + fast lane, hook-enforced phase guards, machine-readable AC gating (acs.json), adversarial QA, tool-restricted subagents
• Session injection: the laws + your live sprint state load at every session start — inert in repos without a harness board
• Windows-safe, never-blocking hooks: polyglot dispatcher, fail-open by design
• Zero network, zero dependencies, zero telemetry — see SECURITY.md
• Deterministic self-update: /capiva:update + /capiva:update-project migrations

Full details: CHANGELOG.md · README · MIT licensed