v0.5.1

What changed

• fix(pipeline): deterministic exact-match greeting fast path (follow-up to #488) (#493)

Full changelog: v0.5.0...v0.5.1

v0.5.0

Stella 0.5.0 is the first minor release of the 0.5.x line — it gathers the stabilization and feature work that accumulated across the 0.4.x patch series into one coherent step: a smarter edit path, graph-aware tooling, a durable context engine, and fleet-scale telemetry.

Editing & tools

• read→edit drift oracle (#331) — edit_file records the content the model last saw and, on a mismatch, distinguishes an out-of-band file change from a genuine not-found, echoes the current content, and lets the model recover instead of retrying a stale edit.
• apply_edits — transactional multi-file edits (#333) — apply a batch of edits across many files atomically, with a dry-run preflight that reports exactly what would change before anything is written.
• run_tests scope:"impacted" (#334) — graph-driven test selection: walk the code graph from your changes to the tests that exercise them.
• graph-derived planner context (#342) — deterministic goal-path localization gives each turn a tighter working set.

Context engine

• Adaptive-context baseline, Phase 0 & 1 (#420, #423) — a typed record taxonomy with scope and temporal axes plus canonical JCS hashing for stable content identity.
• Typed decision events + policy-plane journal bridge (#364) — context receipts §6.3/§6.4 emit loop-detection, budget, retry, and policy decisions as parseable twins of the agent's prose.
• Compaction preserves the prompt-cache prefix (#372) — dedup keeps the earliest duplicate so the provider's cache prefix survives.

Fleet & telemetry

• Observatory: org / hub telemetry dashboard (#410) — aggregate telemetry across an org's agents and workspaces with hub-and-spoke scoping.
• Benchmark harnesses — loop-bench turn-loop + context-query correctness harness (#417) and the stella arena arena-bench adapter recording ContextGraph traces (#425).

Terminal UI

• Markdown & TOML syntax highlighting in the skills/agents editor (#484).
• Interactive Command Deck with animated terminals on the docs landing page (#485).

Fixes

• Loop detection now sees through read_file's session-tally footer, so legitimate re-reads aren't misread as a loop (#440).
• The conversational fast path no longer authors a witness test for a bare greeting like hi (#488).
• Exhaustive TUI event matches handle SpeculationDiscarded cleanly, unbreaking the clippy gate (#422, #430).

Install & upgrade

brew install macanderson/tap/stella   # or: brew upgrade stella

Prebuilt binaries are attached below for aarch64/x86_64 macOS and Linux — verify a download against SHA256SUMS.

Full release notes: https://stella.oxagen.sh/docs/release-notes

v0.4.92

What changed

• fix(pipeline): conversational fast path — a bare "hi" no longer authors a witness test (#488)
• feat(cli): graph-derived planner context + deterministic goal-path localization (#342) (#438)
• feat(tui): markdown & toml syntax highlighting in skills/agents edit mode (#484)
• docs(site): interactive Command Deck + animated terminals on the landing page (#485)
• chore(release): sync versions to 0.4.90 [skip release] (#487)

Full changelog: v0.4.91...v0.4.92

v0.4.91

What changed

• feat: context receipts §6.3+§6.4 — typed decision events + policy-plane journal bridge (#364) (#437)
• chore(release): sync versions to 0.4.89 [skip release] (#486)
• feat(stella-tools): run_tests scope:"impacted" — graph-driven test-impact selection (#334) (#436)
• feat(observatory): org / hub telemetry dashboard (#410) (#434)

Full changelog: v0.4.90...v0.4.91

v0.4.90

What changed

• docs(design): semantic-resolution bridge evaluation (#335) (#435)
• chore: remove OS junk, dead duplicate source, and mis-nested design docs (#428)
• fix(core): compaction dedup keeps the earliest duplicate to preserve the prompt-cache prefix (#429)

Full changelog: v0.4.89...v0.4.90

v0.4.89

What changed

Full changelog: v0.4.90...v0.4.89

v0.4.88

What changed

• feat(tools): read→edit drift — attribute and correct stale-content edits (#331) (#432)

Full changelog: v0.4.87...v0.4.88

v0.4.87

What changed

• fix(core): loop detection sees through read_file's session-tally footer (#440)

Full changelog: v0.4.86...v0.4.87

v0.4.86

What changed

• fix(tui): drop duplicate SpeculationDiscarded arms shadowed since #422 — unbreak clippy gate (#430)
• feat(cli): stella arena — arena-bench adapter port (contextgraph-trace recorder) (#425)
• feat(context): Phase 1 (installment 1) — record taxonomy, scope, temporal, JCS hashing (#423)
• feat(context): Phase 0 — adaptive-context baseline, characterization & ADRs (#420)
• feat(bench): loop-bench — cheap turn-loop + context-query correctness harness (#417)
• chore(release): sync versions to 0.4.85 [skip release] (#424)
• fix(tui): unbreak main — handle SpeculationDiscarded in stella-tui's three exhaustive event matches (#422)
• chore(release): sync versions to 0.4.84 [skip release] (#419)

Full changelog: v0.4.85...v0.4.86

v0.4.85

What changed

• feat: context receipts §6.2 — compaction identities (which blocks left context) (#401)
• fix(core): surface observed budget Warn on the engine path + session-axis BudgetTick (#418)
• fix(pipeline): handle SpeculationDiscarded in replay event_signature (#421)
• fix(cli): make --budget cap the whole run, not each turn (#360) (#416)
• fix(core): stop speculation firing hooks for uncommitted calls; account for discarded pools (#415)

Full changelog: v0.4.84...v0.4.85