v0.16.0

What's Changed

• Fix #314: run CI on macOS and with -race by @vicgupta in #348
• Fix #319: split telemetry exporter into span, transport and API modules by @vicgupta in #351
• Fix #335: Store memory under ./.tinyagent/memory (cwd only) by @vicgupta in #352
• Document workflows and fix backtick handling in workflow files by @vicgupta in #353
• Fix #336: Read memory.toml and plugin data only from --cwd by @vicgupta in #354
• Fix opencode provider: drop --dir, run in --cwd, accept Zen model names by @vicgupta in #355
• Fix #337: Cache TinyGo toolchains under ./.tinyagent/cache (cwd only) by @vicgupta in #356
• Fix #318: align Makefile with CI and add test-race/lint/fmt targets by @vicgupta in #357
• Release v0.16.0 by @vicgupta in #358

Full Changelog: v0.15.1...v0.16.0

v0.15.1

What's Changed

• Fix chat prompt redrawing a new line on every keystroke by @vicgupta in #349
• Release v0.15.1 by @vicgupta in #350

Full Changelog: v0.15.0...v0.15.1

v0.15.0

What's Changed

• Fix #311: inject model.Runtime instead of package-level provider globals by @vicgupta in #343
• Fix #312: rune-safe truncation with one shared implementation by @vicgupta in #344
• Fix #313: keep README/CONTRIBUTING tool and dependency lists in sync with code by @vicgupta in #345
• chat: render Markdown replies and polish the interactive UI by @vicgupta in #346
• Release v0.15.0 by @vicgupta in #347

Full Changelog: v0.14.0...v0.15.0

v0.14.0

What's Changed

• Fix #310: share one tools.Executor builder across fresh, resumed and chat runs by @vicgupta in #340
• Add opencode CLI handoff provider by @vicgupta in #341
• Release v0.14.0 by @vicgupta in #342

Full Changelog: v0.13.0...v0.14.0

v0.13.0

What's Changed

• Refactor ACP session setup into shared session code (prep for chat) by @vicgupta in #301
• Add interactive chat mode (tinyagent chat, -i) by @vicgupta in #302
• Fix #277: tinyagent triage ranks open issues into a next-up list by @vicgupta in #303
• Fix #278: --workflow-observe proposes project rules from failed workflow runs by @vicgupta in #305
• Fix #281: contract docs for stable surfaces in skill workflows by @vicgupta in #315
• Fix #282: document the 256 KiB MCP cap for --tool-output-files by @vicgupta in #316
• Fix #283: Add short PR body template by @vicgupta in #320
• Fix #284: verify-on-stop on by default for --workflow runs by @vicgupta in #321
• Fix #306: one tools.Selection owns the catalog and its resume fingerprint by @vicgupta in #322
• Fix #307: collapse Executor.Run* wrappers to a single RunResult entry point by @vicgupta in #323
• Fix #308: Add agent.New(Options) to validate construction invariants by @vicgupta in #324
• Fix #309: add tools.Executor.Child(ChildSpec) for child executors by @vicgupta in #325
• Part of #310: move resume flag-versus-checkpoint checks into internal/resume by @vicgupta in #333
• Part of #310: fold root, fingerprint, and tool-name checks into resume.Check by @vicgupta in #334
• Part of #310: add resume.Load and resume.Apply for resume assembly by @vicgupta in #338
• Release v0.13.0 by @vicgupta in #339

Full Changelog: v0.12.0...v0.13.0

v0.12.0

What's Changed

• Fix #274: add verify-behavior workflow stage that checks observed output by @vicgupta in #286
• Add claude-code and codex-exec CLI handoff providers by @vicgupta in #287
• Blocked workflow status: park tasks whose checks can't be met instead of failing the run by @vicgupta in #289
• Spec gate: re-plan once, then surface tasks lacking acceptance criteria or a runnable verify command by @vicgupta in #288
• Add tinyagent skills validate for skill-workflow files by @vicgupta in #290
• Read config from ./tinyagent.yaml in --cwd; add tinyagent init and --trust-config by @vicgupta in #294
• Respect NO_COLOR and add --color=auto|always|never by @vicgupta in #295
• Better --check output: summary table and --check --json by @vicgupta in #296
• Error UX: next-step hints for known setup failures by @vicgupta in #297
• Release v0.12.0 by @vicgupta in #298

Full Changelog: v0.11.1...v0.12.0

v0.11.1

What's Changed

• Docs: make the workflow file format unambiguous by @vicgupta in #271
• Docs: subagents are read-only researchers, workers are read-write builders by @vicgupta in #272
• Release v0.11.1 by @vicgupta in #273

Full Changelog: v0.11.0...v0.11.1

v0.11.0

What's Changed

• Docs: cover verify-before-stop, token budgets and help changes since the docs split by @vicgupta in #253
• Fix #249: project rules (ratchet) from agent failures with incident and expiry by @vicgupta in #254
• Part of #250: computational checks for verify-on-stop by @vicgupta in #255
• Fix #250: inferential review tier for verify-on-stop by @vicgupta in #256
• Fix #251: session handoff file for long-running work by @vicgupta in #257
• Update: step-by-step progress and # download bar by @vicgupta in #258
• Help: show task, model and common options in brief help by @vicgupta in #259
• Fix #260: GitHub Action eval input with pass-rate gating by @vicgupta in #261
• Model routing 1/5: role resolver (no behavior change) by @vicgupta in #262
• Model routing 2/5: auxiliary role for compaction and the verify review by @vicgupta in #263
• Model routing 3/5: delegated role with fallback chains by @vicgupta in #264
• Model routing 4/5: task-based routes and --no-routing by @vicgupta in #265
• Model routing 5/5: usage_by_role by @vicgupta in #266
• Docs: group the README documentation table by topic by @vicgupta in #267
• Docs: top-level Tasks and Models pages by @vicgupta in #268
• Sandbox: Bubblewrap engine with Landlock stacked inside (--sandbox-backend) by @vicgupta in #269
• Release v0.11.0 by @vicgupta in #270

Full Changelog: v0.10.3...v0.11.0

v0.10.3

Features

• Token budget (#246, part of #236): --token-limit n / config token_limit caps provider-reported input+output tokens for the whole run. At the limit the run stops cleanly with the new status TokenLimitExceeded (exit 1, resumable; the printed --resume command doubles the limit). Delegates and workers get only the tokens left in the parent's budget, and a worker batch shares one token gate. Off by default (0 = unlimited).
• 80% budget warning (#246): when a token or cost budget reaches 80%, one "Budget notice" user message is added before the next request and a budget.warning event goes to the JSON stream, log DB and OTel trace.
• Token limit everywhere (#247, closes #236): --check shows the token limit on the limits line, --init-config includes a commented token_limit: 0, --eval suite tasks accept limits.token_limit, and --workflow accepts token_limit in defaults and per stage.

Changes

• Behavior change for --attempts (#247): --token-limit is now one budget shared by all attempts, like --cost-limit and --time-limit. Each attempt gets the tokens left, no new attempt starts once the budget is spent, and an attempt ending TokenLimitExceeded stops the run with budget exhausted. Previously each attempt got the full limit.
• The turn cap (--step-limit) stays off by default. A top-level --token-limit with --eval is still a per-trial cap.

---

What's Changed

• Part of #236: token budget and budget warning by @vicgupta in #246
• Fix #236: show token limit in --check/init-config and forward it through eval, attempts and workflows by @vicgupta in #247
• Release v0.10.3 by @vicgupta in #248

Full Changelog: v0.10.2...v0.10.3

v0.10.2

Changes

• A bare subcommand (tinyagent sandbox, memory, planner, plugins, tinygo) at an interactive terminal now prints the same help as tinyagent <sub> --help and exits 0 (#244), for every registered subcommand. Previously sandbox, memory and planner failed with error: --task is required. With stdin piped or redirected they keep the old behavior, so scripts are unaffected; actions, flags and prompts that start with a subcommand word are unchanged.

---

What's Changed

• Show subcommand help for a bare subcommand by @vicgupta in #244
• Release v0.10.2 by @vicgupta in #245

Full Changelog: v0.10.1...v0.10.2