VikingOwl91/gnoma

Changelog

• d5958203cb262e155dcccc285cef45de0965c9ff Merge pull request 'chore(todo): mark post-audit security work complete' (#5) from chore/todo-post-audit-cleanup into main
• 9b9a009f68cb76eb32a94304af4d3a57baf66233 Merge pull request 'docs(security): ADR-004 PostToolUse hook ordering + invariant test' (#4) from docs/adr-posttooluse-hook-ordering into main
• 2132b265a1b14de43295118d58047de0a91e0812 Merge pull request 'feat(security): SafeProvider boundary wrapper (Wave 1)' (#1) from feat/security-wave1-safeprovider into main
• c7eb14ee31c22aed52c6d5e763fc4cf5eb95ce1e Merge pull request 'feat(security): incognito coherence (Wave 2)' (#2) from feat/security-wave2-incognito into main
• dd16c1502917615339a2a72d7a9893c978af8ff7 build(release): publish multi-arch Docker images to GHCR
• bb7892c0c2e27532221006da5dd6c7021f30d037 chore(audit): polish remaining audit findings (M2, H1, H3)
• 13b2f5e14d3e3b5190b357748496086dc455e6d9 chore(lint): clear dead code and tighten lifecycle errcheck
• ec9433d783ea8595f58236e84b87899245f7dca9 chore(lint): clear remaining errcheck and staticcheck findings
• c4fde583f5f045ca5a601b28e63a8fe4421f10a2 chore(lint): gofmt sweep + errcheck cleanups in router discovery
• 343b0fb94ab19c696476c32733ef460326254ca1 chore(todo): mark post-audit security work complete
• 21da29e73e530abc8fe17e9048a9c497638b8682 docs(plan): capture post-SLM-unlock outstanding work
• f8c85a26e9108a3cf10b098384418c1f947d8e34 docs(security): ADR-004 PostToolUse hook ordering + invariant test
• 3ae40083f19e55c661bf47f890dea1e260369f03 docs(security): Wave 2 plan — incognito coherence
• 3a638cc9a4edbc0df659a2a6dfcfa889d45145b8 docs(todo): add stream-error failover to backlog
• 7c6291ab7eb05a1a7a87f3bd36bbd79dcb7a7d2c docs(todo): mark Phase C complete in post-SLM plan summary
• 8b2202e8ec0ac5f7438d7587cb2795faaafcd489 feat(classifier): Wave A — TaskClassifier interface + HeuristicClassifier
• 71f31559c2bf99248ccc94f90549b4f2e1dd7096 feat(cli): add 'gnoma providers' subcommand
• 8450005b31e64471cc730b8610ed105f3df671eb feat(cli): gnoma profile list/show subcommands (Phase C-2)
• 635dad660cfb1a00aaea9706693e0b6cf7f814df feat(config): per-profile config layering with --profile flag (Phase C-1)
• 9fb520fba626c1fab03832f5948f9c80bfdf9e71 feat(engine): M8 cleanup — Wave A wiring gaps
• 176926924c5d2da82e86fd67b6a570abefd90c4e feat(engine): M8 cleanup — Wave B skill enforcement
• aca830e7dbc181d6a66733740b618cce6950abf7 feat(engine): consumption-time stream-error failover
• 397a39250cc1cacbb8b9f8a395018a376b015727 feat(engine): early-stop detection for runaway agent loops
• 43ea2e562d98ab1da8fffa295d53cde34ed11228 feat(engine): two-stage tool routing for small local arms
• b60aa02bfdfc942a6a75f800565c29b4661bebfb feat(fs): enforce workspace boundary on fs tools
• 6c47f8643b76c1ea3b73349f13ebff53d75c381c feat(m8): MCP client, tool replaceability, and plugin system
• 9388479b03e608f1fb866a3adf8a5aa374c7d103 feat(openai): lexical repair for malformed tool-call arguments
• dc438ea1814bb9a92075fb9c1c3e59c67798e9ce feat(plugin): trust-on-first-use manifest pinning
• 44d0bdc032a1ed80a41c236f2d727e975a16f412 feat(provider): subprocess CLI provider for claude, gemini, vibe
• 0b1392cf6be90caf19dceea53674b7e67c9e30bd feat(pty): Phase 2 — interactive shell and bash interactive detection
• 58beb7ce3cc01ec5251d6c2f9c5eba3299498592 feat(router): classifier-source telemetry + router stats command
• 7fbb5454ee00e0c16c518400ed89e0bb1705fb32 feat(router): normalize effort/thinking abstraction across providers
• 0aabd199068247e95ab4e41a140c883b18f9346e feat(router): per-arm strengths + cost weight (Phase D)
• 6883c2a041fa1b885462eba8cd3664927ba9d530 feat(router): tier-based routing — CLI > local > API, disabled arms
• 8dcca64e41d5beaba923aa278115d9551a8c2776 feat(security): add SafeProvider boundary wrapper (W1-1)
• 3c875276c9d1b0e89f3106faf58f1a28d5b7f464 feat(security): implement multi-wave audit remediation and agy provider support
• 34f6f1c786bf9f8b7df784779b1a843d65c3adbe feat(security): incognito coherence across firewall/router/persist (Wave 2)
• dc084d5a82763de44b0b84f58a9a12c6a44078e2 feat(security): wire SafeProvider into all provider sites (W1-2/3/4)
• d6614545a9d26ffb00c44f160f9a8eb199684e96 feat(security): wrap engine.Config.Provider + SetProvider doc (W1 follow-up)
• 893880039b4892f01b5e79b2bf94b5a8ca41155f feat(skill): TUI integration — /skillname invokes skills, /skills lists them
• 61adb24773fae8b36bfdbc187ad439f526af7f72 feat(skill): bundled /batch skill with go:embed
• edc0e97efc2c7c2cb3dafa867d034a6e925249e4 feat(skill): core Skill type and YAML frontmatter parser
• c07ec63419f17f3983e2df21c3fc30920756eebd feat(skill): enhanced coordinator prompt with fan-out and concurrency guidance
• 48c7b7aad4333146ba7f995185f2820fce863e76 feat(skill): pipe mode support and main.go wiring
• b60daf99407e2cdfb67431e26cc3434262a7fafe feat(skill): registry with multi-directory loading and precedence
• ead91e6ccf6fda8064699fb7f64c95fa49f2c7b1 feat(skill): template rendering with Go text/template
• d1a5c79fa4c382ccc8f55601b9951dff1a8b64c8 feat(slm): Wave B — Manager, Manifest, download, subprocess lifecycle
• a9213ec3827db55455ff280eda12bf7e8d36a481 feat(slm): Wave C — SLM classifier, MaxComplexity routing, CLI subcommands, TUI status
• a14fe8b5043e6511512b9589b6ee32d9a3ee08dd feat(slm): pluggable backends + trivial-prompt routing
• b331dcd61a8952197b4ad461d1820bffdc178720 feat(subprocess): per-agent binary override via [cli_agents] config
• f8867f5d7866973d9e12c1c648d089a36ced096c feat(tui): /config opens interactive settings panel
• d84b295da22c503b7bffaa093df86c4c7f00bd3f feat(tui): /profile slash command + status-bar profile badge (Phase C-3)
• 48e63a9bc08afde3c841234a7ac6324680443905 feat(tui): Tier 1-2 UX improvements — completions, usage, provider status
• ce5f9d3dc95a781e90e0c34d10273ded0c6774e1 feat(tui): Tier 3-4 UX improvements — split, routing, session naming, context bar
• befcbdcfefcae2dc84a4e62e384b0d04ce2091a2 feat(tui): suggestion box above input, input mode indicator, ! execute
• 0b4de6054d91c0dfd2969a62e385d004d282e97f feat(tui): surface SLM backend + per-turn classifier in status bar
• 7a417b9011d6d719dfce996f2ba5e0a1e7d23b32 feat: /config, /model listing, /shell stub, /help update
• 167db19bfbd89f316381d8ff1cc207b14dd66798 feat: /resume TUI command + SessionStore in tui.Config
• 50bb5f2f6b3d029d88657e166ce1bfb5f5d75a4a feat: AgentExecutor — elf-based hook evaluation via elf.Manager
• 54df0f2f08f7f425d0b6aa3dbd02f4f3e5837d4b feat: CommandExecutor — shell hook execution with stdin/stdout protocol
• 47ae6775f3b91662b18c4765a36f0edcb7e787fd feat: Dispatcher — handler chain dispatch, filtering, transform chaining
• b421439087c081122c7cbe88fdf7a1be0d8a03a2 feat: Engine.SetHistory/SetUsage/SetActivatedTools for session restore
• 3c5aa3639aed52f48ba869592e7eb72d1bbef809 feat: JSON serialization for Message and Content (session persistence blocker)
• 2f60bd9f0a8bb374fdb5a1cae5d3178baa7e7378 feat: LocalConfig + auto-save hook in session.Local
• 11363f3b976fce2ea2e246f51a0ffa8eccce0144 feat: M1-M7 gap audit phase 2 — security, TUI, context, router feedback
• 95dfd0cf0ceac1c02fa0e17e066b3481e6b3b98b feat: M1-M7 gap audit phase 3 — context prefix, deferred tools, compact hooks
• 63f4c1389e82977ab47291eceeb6746b7ca797fd feat: M6 complete — summarize strategy + tool result persistence
• 704f3a7302762200f2f187f7a73382ef49e50d05 feat: M6 context intelligence — token tracker + truncation compaction
• 13db7521b112478ac6bdf3955544e0dcbce5628b feat: M7 Elfs — sub-agents with router-integrated spawning
• 4f1e0cf567a8593b5e60c5adf4ee96ebdabe5ced feat: Ollama/gemma4 compat — /init flow, stream filter, safety fixes
• 685e3b97f22427bb613be78433930ebfccd94a8d feat: ParseHookDefs — config to HookDef conversion with validation
• 45c0d0c43e0f537997523f8996e654793d3c76fe feat: PromptExecutor — LLM-based hook evaluation via router
• 64ee385039a88bda43491d15056419b490163041 feat: QualityTracker — EMA router feedback from elf outcomes, ResultFilePaths tracking
• 39181168b69971d95059107d0777a7454ad4a94d feat: QualityTracker.Snapshot/Restore + Router.QualityTracker() for cross-session persistence
• 9d18f1179a353aa1d6f1476382149968355069af feat: SessionStore — save/load/list/prune session snapshots to .gnoma/sessions/
• 96a336aa0e32938373fec25b1994e7af19111f47 feat: TUI slash commands, incognito toggle, model switching
• 350b7bbe05fa3a39583323c556a0962823546244 feat: accurate context window sizing from arm capabilities + prefix token baseline + tokenizer wiring
• 9e7caf2467646911b356322f860f394457986a14 feat: add Anthropic provider adapter
• f3ca791122903bbb2c4a3d78baf1f0ca37215efe feat: add Bubble Tea TUI with interactive chat
• dccb5fe65a0cb2815f77f0196d82ab77406670c3 feat: add Google GenAI provider adapter
• 261c19f90f5d6f21118ff953ff3607b160933f03 feat: add OpenAI provider adapter
• 9608436b52ce5ca1d1b0dbc7ff6b768b00f00e47 feat: add OpenAI-compat adapter for Ollama and llama.cpp
• a7d86054de4774875497284370568dc9746cdf34 feat: add Session config section (max_keep for session retention)
• 291ed3534134a8d8cb5ad7c0dbf6c530f2def5e6 feat: add TOML config system with layered loading
• 17d83f2e2abf77f2b2795f8bb0902ddb15ad7e83 feat: add agy CLI provider and support structured output via prompt augmentation
• 0d2d825e5278ae59094657c15c922cc8c7bbd30d feat: add dynamic model discovery within providers
• 788bd8ec24b3a6648d8ed5cf6c886e6d83f6aac5 feat: add foundation types, streaming, and provider interface
• 03c470bf651f7dc6f883631ff35a5eeb49bd559c feat: add permission system with 6 modes and compound command parsing
• 847735a9f7ee594470fbfe76b4246227913de4f0 feat: add router foundation with task classification and arm selection
• 09f102bdec9427235d662e603d6a7f2296b2d37a feat: add security firewall with secret scanning and incognito mode
• 0f93ee18e415efa7f53b25c74209a142ec0cb5a2 feat: add session interface with channel-based local implementation
• 60883521c72a730ea90488283233b2de0459d337 feat: auto permission mode, edit diffs, truncated tool output
• 76916846aaa39bb5345160f485b3fe4d9aa3312f feat: auto-discover local models from ollama + llama.cpp
• 8f4a4672e7cfd4b7d947191dcf4b5c6178c72adf feat: colored permission mode indicators on separator lines
• 6cfe35620dc07dd7f6be417b758e760385dfc1ef feat: compact system inventory with queryable system_info tool
• 46505a1f71b1fd8ead79794a42180c1706d3d5ff feat: complete 7/7 bash security checks
• 69f5dba0916a67b25d0d41f7ef1004d077559af8 feat: complete M1 — core engine with Mistral provider
• ebfbefc73d81f72fb49c020d6427d37f379550ee feat: configurable max_turns for elfs — LLM sets via agent tool param
• f7a22287657282a700995f767f3b3e80b2fd5762 feat: coordinator mode — system prompt injection for orchestration tasks
• 26666e6d2ca3a592e2362c3490910954baa4bb80 feat: coordinator mode — system prompt injection for orchestration tasks
• bb93317fb610ac11e443850b837768caf7db0214 feat: ctrl+o toggles tool output expand, fix auto default
• 1ec90b0ad746713a2f12d6aad8a7145bff4e33b6 feat: engine hook integration — PreToolUse, PostToolUse, Stop
• 1f620d2725601d75731f775806415baf972c1bbd feat: hook config schema with user+project merge ordering
• ea7ec98d76c3b4210dfa7c5b4f395ed70256d622 feat: hook core types — EventType, Action, CommandType, HookDef, Executor
• 0297f56d8f65cd16f42d867cb6e27a72ad47868f feat: hook payload marshal/unmarshal helpers
• 8e5ddb20cb7964233bc72f38ac94efa38763ca81 feat: hybrid system inventory — dynamic PATH scan + runtime probing
• 9a78af7b059f3305f60db3eaf57335b2b4ac6541 feat: inject mode changes into engine conversation history
• c0631728f697a6c64b18a56413b0fa2aa8c7e4e7 feat: interactive permission prompts in TUI
• 12ace89e31a84724ae3d0836a1b8ad7e3e7277e5 feat: interactive session picker for /resume and --resume
• 745b27e5db7b8008744ddeb3062eca66c3a0d18a feat: list_results + read_result tools for coordinator artifact discovery
• f4fda8346b41ce209a942c8f80fc5a34d2c6834b feat: list_results + read_result tools for coordinator artifact discovery
• c01069164e9d39b05a118f301c30800be8ca58c6 feat: live elf progress in TUI
• d71bd942c4464d14274ffaf7bd99cf3633c2c3f6 feat: local model reliability — SDK retries, capability probing, init skill, context compaction
• 7c96f6a29185bf809212bf62451833da3dd383f6 feat: markdown rendering in chat via glamour
• 84d9e636b741f553e3e2970eeb997b6254ad76b5 feat: multiline input with auto-expanding textarea
• 4ca7eadd38296b35e1576ff07a0542972a4aa23d feat: permission mode switching in TUI
• 6fa9df5613341388c989c7106e3df44dfe5eaa94 feat: persist.Store — session-scoped /tmp tool result persistence
• e1a47a762089fc4a0bf0b2c87b6b3d146bf0b40b feat: rate limit pools, elf tree view, permission prompts, dep updates
• abb3e3ca90698d317fc2680c5843bf5624f5cfa8 feat: spawn_elfs batch tool for guaranteed parallel elf execution
• 6a0a6543346a5a107f85450b8428ddfcc31fdd7a feat: tiktoken tokenizer — accurate BPE token counting with provider-aware encoding
• 9a0e02e1c6b263bcb1d01d1f0da56e4f65c836a4 feat: tokenizer-aware Tracker.CountTokens/CountMessages replaces EstimateMessages in compaction
• 135c8afe8057e8d433ce9cd2fab6fccba7ef0275 feat: various improvements to engine, router, and TUI
• 4596ea21562a9870f81ac9bd5efd77872736b1fd feat: wire --resume/-r CLI flags, SessionStore, quality persistence
• 86aabd4946238bbbc58e9ee30f0d6cd1f315255b feat: wire config system into main — TOML config now active
• 8d9c521f7a46776de80e5aa1fe320fbd9e26a8d5 feat: wire hook dispatcher in main.go — SessionStart, SessionEnd, PreCompact
• 24b5126d66f58621078d7c79fd675171dcc182e6 feat: wire permission checker into engine tool execution
• dae2c488e5b5820558c6ee20f6b9515dcf8ac1dd feat: wire persist.Store into engine, elf manager, and agent tools
• 062566a23de1a861bb0b33f8ae42b656d047cdef fix(cli): three UX issues — help output, TUI startup, setup command
• 5cd3ccd9318d2f69606bb1d6b550383ab0712c2e fix(engine): guard mutable state with a mutex
• c44db99b41dc07c5e827f9a9b81c0c40edeaaead fix(hook): execute hook Exec as a binary, not via sh -c
• 6bb9c33d04bf427e2c39cd1ff642ec27b67cb767 fix(m8): replace_default map, error UX, benchmarks, and launch prep
• 67948df8cb998575f8ff2bf37737746d5b18e544 fix(mcp): make transport cross-compile on Windows
• 59d7e2fa43d614338da6b11a3e0341d26f0c0744 fix(release): target GitHub mirror; provider-neutral README; correct NOTICE
• 0caab0fed12417cb56801e3a4762d7f43b35fff1 fix(router): discovery loop removes forced arm, breaking routing
• 8539426a467c37d1ebbe08a12c1b7be4da55b3c5 fix(router): restore Ollama cache prune + provider-specific context defaults
• f6f8801040177f51d377e0a1b9b5de35e1216c85 fix(router): restore llama.cpp model enumeration; keep /props for n_ctx
• eb0583f606d052b1db0f5a0713c55e1117557e07 fix(router): unpin config-default provider + complexity floor by task type
• 8f13ed78a9e8cc8e7f334c483559a9be7a85953d fix(security): redact truncated private keys via header-fallback pattern
• 329610209a239ac32be314e3ce2438f45b5a729c fix(slm): invoke llamafile via sh to bypass Wine binfmt_misc
• 9037a0d195614128aceca78ce566862f414ba530 fix(slm): skip re-download when already set up
• adb4f5db5d7b58b31775649ba371c79c311ff945 fix(slm): start llamafile in background; use lazyClassifier
• c8813768d5dc9b44d94cd1c3cd46d3f355558364 fix(subprocess): harden agy CLI integration
• 07a976c32a10d4526c4bde1e1816dc48c5d84767 fix: ClassifyTask priority ordering — orchestration below operational types
• de1798ff5cd93023e153ef6796e18cdecf22b7f8 fix: M1-M7 gap audit phase 1 — bug fix + 5 quick wins
• 5b14b0ac84127898f250f6a1aee097618a2267dc fix: TUI overflow, scrollable header, tool output, git branch
• 90f1fc872b5e4e5345855e830737debb40059441 fix: TUI spacebar + improved design
• e04cacc21557dde17f0154b08d879813cf3ccbc5 fix: append mutation, pipe-mode hang, Mistral regex false positives
• 92d2921ea170d3331271356a9d9b5f3778473b84 fix: consistent indentation and AI icon in chat
• 302dd2ac1d98e417a32df1eb55a54e1c624b1a8f fix: default permission mode, incognito shortcut + display
• 2093beea5886dcbd8957676b677fa3dbe98586a0 fix: deterministic 500 retry, OpenAI error wrapping, local /init prompt
• 086a4622fdb77cb1d337da53576279fc8226bf8f fix: elf progress — proper last-2-lines tracking, 70 char truncation
• 8138bd69f8bc94ea1bbd57c5ca0d893b41a59490 fix: live elf progress shows tool calls + results, not just text
• 0f8bf7dbdd1230e57ff6001500c94b39125a4cd4 fix: mode change messages tell model to retry
• 88e76cddb0fcc769d15387d8772b4dcdd4819c8a fix: persist.Store — sanitize callID, log save errors, document List filter semantics
• 0a1730943f1696796ab9cff2ed6cbc48117b8954 fix: provider-agnostic startup + slm setup auto-config
• 3bdd9cec7e7e00c65bc6a32be3a93f0171d6390e fix: raw text during streaming, glamour only on completed messages
• 801036a793495fc7e6b841cb769d67057e29c136 fix: restore import order in main.go
• 6aea2a9e3a95ba0f43c78450e4dac568a22ec227 fix: retry with exponential backoff on 429, stagger elf spawns
• ae9683818ba08de34fede52f8499c3e91fe66ba2 fix: session security and correctness — path traversal, turn count restore, incognito quality leak
• aaf3c1e42cee4fdcb2f886fa0f74e05556e70385 fix: strings.Builder panic — use pointer to avoid copy-by-value
• d2139c6f0c2cb488f4a70e7f90916eaf649e1373 perf+feat: parallel startup discovery + slash-command suggestion dropdown
• 99fa0ff08eb3d73ea45de1b7391b3a2ded720d18 refactor(providers): refresh defaults to current 2026 model lineup
• 9853a522e6f85703570d34df81d2de80eda38054 refactor(security): consolidate TOCTOU-safe path canonicalization
• fb422028349ce19adfefd8d7a0cec436907993a2 refactor(security): seal SecureProvider via unexported marker method
• c54471a37be131b9f21485662bb7d192154b82d0 refactor: migrate mistral sdk to github.com/VikingOwl91/mistral-go-sdk
• 2bf700eec2ffe3b3f4bec909ec72a6013f7dca4b test(elf): make mockProvider.calls atomic
• beaa09a154684325800de969c14f8ef86a31ea8a test(permission): lock in elf safety-pattern inheritance
• 342b3903e1c5d74ae33f746628b2e2ee81054e14 test(slm): align HappyPath with task-type complexity floor