v0.2.0

0.2.0 (2026-10-22)

Features

• cli: add export --format skill (#111) (62c68b8)
• cli: add report --final (#110) (bce0280)
• cli: add the fit command (#104) (e2048cc)
• cli: add the prune command (#106) (41fe8c2)
• eval: add batch and held-out gates and the tie-break (#94) (dd70d0d)
• eval: add report statistics and the delegation map (#107) (8669cac)
• eval: log control evaluations and freeze the harness (#109) (a12a88b)
• harness: validate and apply proposals (#95) (8eb1a43)
• loops: ablate accepted changes and prune the useless (#105) (1d6fe40)
• loops: add the middle loop round (#103) (96b2f85)
• providers: add the Messages API provider for the tutor (#97) (2e7050a)
• reports: report progress since the baseline (#108) (9f62d56)
• storage: add change records for the library (#102) (2c14ada)
• tutor: diagnose failed runs into one proposal per call (#98) (8a3739f)
• tutor: record a reflection on each decision in EXPERIENCE.md (#99) (d9590ee)

Bug Fixes

• report: skip run records cut off by a crash (#112) (a3911f9)

Documentation

• adr: free-text harness files are content, not formats (#101) (6535d50)

v0.1.3

0.1.3 (2026-10-21)

Features

• config: add prompt-cache prices for the tutor (#92) (0ba3bcb), closes #91

v0.1.2

0.1.2 (2026-10-21)

Bug Fixes

• config: default to the sandbox image of the installed version (#88) (1f3980b)

Documentation

• adr: a new default for a written-out field is not a format change (#89) (dd3ac25)

v0.1.1

0.1.1 (2026-10-21)

Bug Fixes

• ci: publish the sandbox image for arm64 as well (#85) (3a5bf93)
• sandbox: build per-repository environment images (#84) (b3343b7)

v0.1.0

0.1.0 (2026-10-21)

Features

• cli: add baseline and report (#82) (5a90709)
• cli: add doctor with speed, tool-call probes and tier (#77) (93e9409)
• cli: add init and tasks mine, list and show (#79) (d2c5ba5)
• config: add saddlefit.yaml model and loader (#55) (1a9c31a)
• context: assemble the harness context within budget (#71) (4d741b5)
• eval: add failure taxonomy and tamper detection (#64) (0a7c3b4)
• eval: add leak guard that aborts on canaries (#65) (91ede13)
• eval: add the evaluator with tamper checks (#75) (e9b4297)
• harness: add git-backed harness store (#74) (29005fd)
• harness: add loop.yaml schema, clamping and allowlist (#67) (fdf8a94)
• logs: add JSON-line logging with secret redaction (#58) (22f9594)
• providers: add ModelClient protocol and JSON tool parsing (#57) (2cbf36c)
• providers: add OpenAI-compatible student provider (#60) (79eb478)
• providers: add retries and scripted fake provider (#59) (0a40318)
• providers: pass reasoning effort to the runner (#80) (e41765b)
• sandbox: add container lifecycle with B11 limits (#61) (2714dab)
• storage: add run records and schema migrations (#66) (1f0a37b)
• student: add the agent loop with finish gate and loop detection (#76) (d713721)
• tasks: add deterministic splits and the suite lock (#62) (8002fd5)
• tasks: add sealed store, canaries and import boundary (#63) (1e61e84)
• tasks: mine viable tasks from git history (#69) (8719e13)
• tools: add student tools that run in the sandbox (#70) (306eedb)

Documentation

• add build specification (1f9c25f)
• add contributor, conduct and security docs (6e8812a)
• add readme, architecture, ADR process and notes (1cb2581)
• add the Saddlefit logo to the README (#81) (b2a27da)
• adr: define what reading sealed data means for evaluation (#78) (b13b7af)
• readme: add loop and landscape diagrams (#68) (f56be57)