brew update && brew upgrade agentops · bash <(curl -fsSL https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install.sh) · checksums · verify provenance

Release v2.20.0

Highlights

• The knowledge flywheel now closes its own loop — learnings are extracted automatically at session boundaries, bridged from handoffs, and promoted to MEMORY.md without manual intervention.
• Production RPI orchestration (ao rpi serve) brings SSE streaming, auto mode, and phased execution to goal-driven workflows.
• Context declarations let skills declare what knowledge they need — ao inject --for <skill> reads frontmatter and retrieves scoped context.
• Knowledge mining with ao mine and ao defrag automates codebase intelligence gathering and deduplication.
• Behavioral guardrails ship as a 3-layer hook system (intent echo, research loop detection, task validation gate) with warn-then-fail ratchet progression.

What's New

• ao session close --auto-extract produces lightweight learnings and auto-handoff when sessions end.
• ao handoff bridges decisions into .agents/learnings/ automatically — no more lost context between sessions.
• ao inject --session-type boosts learnings matching the current session context (career, debug, research, brainstorm).
• ao inject --profile surfaces the durable identity artifact (.agents/profile.md) in every session.
• ao mine and ao defrag for automated knowledge mining and deduplication.
• ao rpi serve <goal> with SSE streaming and auto execution mode.
• ao inject --for <skill> reads skill context declarations for scoped retrieval.
• Coverage ratchet gate, fuzz targets, property tests, and golden file contracts harden the test suite.
• Release cadence gate warns when releasing too frequently.

All Changes (Condensed)

Added

• Flywheel loop closure: auto-extraction, handoff bridge, MEMORY.md promotion, session-type scoring.
• Context declarations with sections allowlist and artifact directory support.
• RPI orchestration engine, headless runner, and knowledge mining pipeline.
• Behavioral guardrails (3-layer hooks), git worker guard, embedded parity gate, swarm evidence validation.
• Nightly CI athena job, BATS integration tests, fuzz targets, golden file contracts.

Changed

• Coverage floor raised to 84% (cmd/ao), average to 95%; complexity ceiling tightened to 20.
• Session-start defaults to lean mode; injection gets hard quality gate (maturity + utility filter).
• Post-mortem redesigned as knowledge lifecycle processor; legacy RPI orchestrator retired (-1,121 lines).
• 10k LOC of coverage-padding tests deleted; 72 stale tests quarantined.
• Skill hardening with web security controls, CSRF protection, and crank pre-flight.

Fixed

• Flywheel signal chain (confidence decay, ordering, glob errors), path traversal in context hook, handoff race condition.
• ao mine stability (dedup IDs, error propagation, --since window, empty output guard).
• Hook tests aligned with warn-then-fail ratchet (strict env required); pre-mortem gate exit code corrected.
• RPI serve pipeline hardened, jq injection fix, goals parser edge cases, rune-safe UTF-8 truncation.

Full Changelog: v2.19.3...v2.20.0

brew update && brew upgrade agentops · bash <(curl -fsSL https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install.sh) · checksums · verify provenance

Changed

• README highlights ao search (built on CASS) — indexes all chat sessions from every runtime unconditionally; adds Second Brain + Obsidian vault section with Smart Connections local/GPU embeddings and MCP semantic retrieval

Full Changelog: v2.19.2...v2.19.3

brew update && brew upgrade agentops · bash <(curl -fsSL https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install.sh) · checksums · verify provenance

Fixed

• CHANGELOG retrospectively updated to document all v2.19.1 post-tag commits (skills namespace fixes were shipped but not recorded)

Full Changelog: v2.19.1...v2.19.2

brew update && brew upgrade agentops · bash <(curl -fsSL https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install.sh) · checksums · verify provenance

Fixed

• Quickstart skill rewritten from 275 lines to 68 lines — removes 90-line ASCII diagram and 50-line intent router that caused 3+ minute runtime; now outputs ~8 lines and completes in under 30 seconds
• truncateText edge case: maxLen 1–3 now returns "..."[:maxLen] instead of the original string unchanged
• Dead anti-pattern promotion functions removed from ao maturity (promoteAntiPatternsCmd, filterTransitionsByNewMaturity, displayAntiPatternCandidates, ~99 LOC)
• Windows file-lock and signal support — replace no-op filelock_windows.go with real LockFileEx/UnlockFileEx via kernel32.dll; extract syscall.Flock and syscall.Kill into platform-specific helpers so the binary compiles on Windows without POSIX-only syscalls
• heal.sh Check 7 false positive — script reference integrity check now strips URLs before pattern matching, preventing remote https://…/scripts/foo.sh references from being validated as local files
• Security gate BLOCKED_HIGH — three persistent findings resolved: gosec G118 false positive (context cancel func returned to caller), golangci-lint nolint syntax (space in // nolint: directive), radon double-counting reverse_engineer_rpi.py from skills-codex/ copy

Added

• Spec-consistency gate (scripts/spec-consistency-gate.sh) validates contract files before crank spawns workers
• Command-surface parity gate (scripts/check-cmdao-surface-parity.sh) ensures all CLI leaf commands are tested
• scripts/post-merge-check.sh now validates go mod tidy sync and blocks on symlinks
• scripts/merge-worktrees.sh now propagates file deletions and preserves permissions
• Post-mortem preflight script checks reference file existence before council runs
• Hooks.json preflight validates script existence
• Windows binaries added to GoReleaser and SLSA attestation subject list

Changed

• Coverage floor raised 78% → 80% with CI enforcement gate; Codecov threshold aligned to 75%
• Six truncation functions converted to rune-safe Unicode slicing
• truncateID in pool.go delegates to shared truncateText
• Crank skill invokes spec-consistency gate before spawning workers
• Vibe skill carries forward unconsumed high-severity next-work items as pre-flight context
• Release skill warns on unconsumed high-severity next-work items
• next-work JSONL schema formalized to v1.2
• Skills installation switched from npx skills to native curl installer (bash <(curl -fsSL …/install.sh))
• README updated with 5-command summary, compound effect section, and /vibe breakdown

Full Changelog: v2.19.0...v2.19.1

brew update && brew upgrade agentops · npx skills@latest update · checksums · verify provenance

Release v2.19.0

Highlights

• RPI operations are more observable and steerable with new stream/worker surfaces and stronger nudge flows.
• CLI and hooks are fully aligned on the flat command namespace, reducing drift and stale command usage.
• Reliability hardening from multiple post-mortem waves landed across CLI, RPI, swarm, and docs/skill parity.
• Codex delivery matured with modular skill generation, runtime-specific overrides, and improved install/bootstrap paths.

What's New

• Added ao mind for knowledge graph operations.
• Added RPI operational surfaces: ao rpi stream and ao rpi workers.
• Improved tmux worker visibility and nudge telemetry in phased RPI runs.
• Added Windows build output support in GoReleaser.

All Changes (Condensed)

Added

• New command surfaces for knowledge graph and RPI event/worker observability.
• Codex install/bootstrap improvements, including native ~/.codex/skills handling and one-liner install flow.

Changed

• Completed CLI namespace flattening and aligned hooks/docs/tests around canonical command paths.
• Regenerated codex-native skills with modular layout and codex-specific overrides.
• Hardened CI/release policy enforcement (runtime sections, parity checks, release-e2e checks).
• Refactored high-complexity CLI paths to improve maintainability and reduce risk.

Fixed

• Post-mortem remediation findings for CLI/RPI/swarm reliability and edge cases.
• Flat-namespace delegation/integration regressions in hooks and tests.
• False positives and stale-path hints in heal/doctor workflows.
• Skill/documentation parity and cross-reference drift.

Removed

• Legacy inbox/mail command surface and stale/dead catalog references.

Full Changelog

See CHANGELOG.md for full contributor-facing details.

Full Changelog: v2.18.2...v2.19.0

brew update && brew upgrade agentops · npx skills@latest update · checksums · verify provenance

Fixed

• ao seed now creates .gitignore and storage directories — reuses setupGitProtection, ensureNestedAgentsGitignore, and initStorage from ao init
• ao seed text updated from stale ao inject/ao forge to current MEMORY.md + session hooks paradigm
• MemRL feedback loop closed — ao feedback-loop command wired, ao maturity --recalibrate dry-run guard added
• Quickstart skill updated to reference ao seed and current flywheel docs
• CLI reference regenerated after ao feedback-loop and seed help text changes

Changed

• .agents/ session artifacts removed from git tracking
• PRODUCT.md updated — Olympus section removed, value props and skill tier counts corrected
• GOALS.md coverage directive updated to measured 78.8% (target 85%)

Full Changelog: v2.18.1...v2.18.2

brew update && brew upgrade agentops · npx skills@latest update · checksums · verify provenance

v2.18.1 — Knowledge Flywheel Integrity Hotfix

Addresses 6 findings from a council e2e validation of the knowledge pipeline. The flywheel now prunes, injects, and compounds as advertised.

Highlights

• Flywheel injection fires every session — default startup mode switched from manual (pointer-only) to lean (utility-weighted knowledge injection)
• Empty learnings can no longer game metrics — extract skips empty files, pool ingest rejects stubs, 13 existing stubs removed
• Pruning is no longer dormant — auto-prune enabled by default, eviction and anti-pattern thresholds relaxed to match real data distributions

What's New

Knowledge Pipeline Fixes

• Extract prompt no longer creates "No significant learnings" stub files
• Pool ingest rejects bodies containing the sentinel phrase as a safety net
• ao pool ingest now runs automatically at session end after forge (was manual-only)

Threshold Recalibration

• Anti-pattern detection: harmful_count >= 5 lowered to >= 3 (no learning ever accumulated 5 harmful events)
• Eviction confidence: < 0.2 relaxed to < 0.3 (enables eviction to actually fire)
• Help text synced with code: promotion threshold documented as 0.55 (was incorrectly showing 0.7)

Default Changes

• AGENTOPS_STARTUP_CONTEXT_MODE defaults to lean (was manual)
• AGENTOPS_AUTO_PRUNE defaults to 1 (was 0). Set to 0 to disable.

All Changes

See CHANGELOG.md for the complete list.

Full Changelog: v2.18.0...v2.18.1

brew update && brew upgrade agentops · npx skills@latest update · checksums · verify provenance

v2.18.0 Release Notes

Highlights

• Knowledge plumbing overhaul — New commands (seed, lookup, curate, dedup, contradict, context assemble) connect the knowledge flywheel end-to-end, from planting AgentOps in a fresh repo to assembling context briefings for tasks.
• Memory sync across runtimes — ao memory sync writes structured MEMORY.md with managed block markers so Codex, OpenCode, and other runtimes can read session history without AgentOps hooks.
• 79% cmd/ao test coverage — Five waves of parallel test generation added 500+ tests covering Cobra commands, RunE entry points, output formatters, and edge cases. Cobra pflag state pollution fixed for reliable test isolation.
• Codex-native skill delivery — New skills-codex/ format with install and sync scripts brings AgentOps skills to Codex and other non-Claude runtimes.

What's New

Commands

• ao notebook update — compound MEMORY.md loop merging session insights
• ao memory sync — repo-root MEMORY.md with managed block markers
• ao seed — plant AgentOps with auto-detected templates (go-cli, python-lib, web-app, rust-cli, generic)
• ao lookup — retrieve knowledge artifacts by ID or relevance
• ao constraint family — manage compiled constraints (list, activate, retire, review)
• ao curate family — curation pipeline operations (catalog, verify, status)
• ao dedup / ao contradict — detect duplicates and contradictions in learnings
• ao metrics health — flywheel health metrics (sigma, rho, delta)
• ao context assemble — build 5-section context packet briefings

Knowledge Pipeline

• Work-scoped injection: ao inject --bead <id> boosts learnings tagged with active work
• Predecessor context: ao inject --predecessor <handoff> surfaces handoff context
• Compact index: ao inject --index-only outputs ~200 token index for JIT retrieval
• Global ~/.agents/ tier for cross-repo learning sharing
• Citation-to-utility feedback pipeline in flywheel close-loop

Hooks & Infrastructure

• SessionStart rewritten with 3 startup modes (manual, lean, legacy)
• SessionEnd gates notebook update and memory sync on successful forge
• Platform-specific process-group isolation for goal timeouts (Unix SIGKILL pgid, Windows taskkill /T)
• Constraint compiler hook and Type 3 setup hook template

Testing

• 500+ new cmd/ao tests across 5 parallel waves (64% -> 79.2% coverage)
• Cobra pflag state pollution fix ensures reliable test isolation
• Coverage floor ratcheted to 78%, zero-coverage max to 20

All Changes

See CHANGELOG.md for the complete list of 53 commits in this release.

Full Changelog: v2.17.0...v2.18.0

brew update && brew upgrade agentops · npx skills@latest update · checksums · verify provenance

Highlights

Goals are no longer just a YAML checklist. The new GOALS.md format adds strategic direction — mission statements, north stars, and steerable directives that tell the evolve loop what to work on when nothing else is queued. This directly addresses the 58% idle rate in autonomous evolve sessions.

What's New

• GOALS.md intent layer — A markdown-based goals format with mission, north/anti stars, and directives. Evolve can now act on strategic priorities when gates and beads are empty.
• Six new ao goals commands — init, steer, prune, migrate --to-md, measure --directives, and validate give full lifecycle control over goals.
• Format-aware writeback — ao goals add auto-detects whether your goals file is YAML or Markdown and writes in the correct format.
• RPI fallback fix — When Phase 1 creates a task instead of an epic, ao rpi now falls back to any open issue instead of failing (#50).

All Changes

Added

• Markdown-based goals format (v4) with mission, north/anti stars, and steerable directives
• Interactive goals bootstrap command with non-interactive mode
• Commands to add, remove, and prioritize directives
• Command to remove stale gates referencing missing file paths
• YAML-to-Markdown goals format converter
• JSON output of active directives for evolve consumption
• Goals format validation and directive counting
• Format-aware goals writeback (auto-detects md or yaml)
• Markdown parser library with round-trip rendering and 26 tests
• Goals skill rewritten with OODA-style verbs
• Evolve directive-based cascade for reducing idle cycles

Fixed

• RPI falls back to any open issue when no epic exists
• RPI phased processing test coverage added for core functions

Full changelog

Full Changelog: v2.16.1...v2.17.0

brew update && brew upgrade agentops · npx skills@latest update · checksums · verify provenance

v2.16.0

Highlights

• Evolve idle hardening — 52% of evolve commits in production were idle noise. Now idle cycles aren't committed, stagnation is disk-derived (survives compaction), and a 60-minute circuit breaker stops runaway sessions.
• Evolve --quality mode — New flag inverts the priority cascade: post-mortem findings before goals. Harvested work has 100% first-attempt success rate in production (15/15 items). Quality scoring tracks unconsumed findings.
• 6-phase RPI E2E tests — Full lifecycle validation: gate retries, complexity scaling, phase summaries, promise tags. Catches integration regressions that unit tests miss.
• 8 ao CLI bug fixes — Pool matching, artifact counts, score display, symlink detection, JSON truncation, and more (#41-#48).

What's New

Evolve Skill Overhaul

The evolve skill received three major improvements based on analysis of 179 production cycles (117 commits, 7.5 days of autonomous operation):

• Idle cycles are no longer committed to git, reducing commit noise by ~52%
• Rolling fitness files (fitness-latest.json) replace per-cycle snapshots
• Canonical cycle-history.jsonl schema standardized across all entry types
• New --quality mode prioritizes unconsumed post-mortem findings

CLI Fixes

Eight bugs fixed across ao subcommands, all with issue tracking (#41-#48). Notable: v1 GOALS.yaml now accepted with deprecation warning + migration path.

Skill Maintenance

13 skills received targeted fixes: dead links, stale references, missing catalog entries, vaporware sections removed, fixture paths corrected.

Testing

Fixture-based CLI regression tests and a 6-phase RPI E2E test suite added for comprehensive lifecycle validation.

All Changes

Full Changelog: See CHANGELOG.md

Full Changelog: v2.15.3...v2.16.0