Repository navigation
Releases: damijanc/damaian
Releases · damijanc/damaian
Release list
v0.39.0
What's Changed since v0.38.0
- Close spec 24 repository map and monorepo boundaries (5d6efc7)
- Keep the desktop shell router's debug stack frame under 2 MiB (8ddeee9)
- Document repository roots and record spec 24's acceptance results (6af1d65)
- Show and edit repository roots in the desktop app (f7c776c)
- Run commands and resolve findings in a sub-root (aa33e96)
- cleanup (e191797)
- Render a bounded repository map into model context (bf58e93)
- Plan spec 50 model-initiated clarification (33c8541)
- Persist the repository map and propose validations per root (a659264)
- Let users add and remove project roots in repository config (dc390f5)
- Build the repository map with per-root metadata and commands (11c6403)
- Classify, propose and allow commands per repository root (323be1d)
- Detect project roots from the index (0834ffe)
- Plan spec 24 as a folder spec with a nine-task breakdown (40bd8b0)
- Close spec 31: permission profiles (c2743de)
- Name policy rules in plain words and document permission profiles (ddeaa0b)
- Export and import permission profiles without credential references (fec0ee4)
- task status update (a33e564)
- Close spec 22: document findings and cover persisted-secret redaction (448832a)
- Add the findings panel with dismissal and scoped repair (028b16a)
- Show the attributed effective policy and a profile picker (62c2aa4)
- specs 22 task 10 planning (8c7402f)
- Attribute every effective policy rule to its source (c9881a2)
- Serve findings, status changes and repair requests from the shell (435527c)
- specs 22 updates (1677e18)
- Build scoped repair requests from selected findings (415d633)
- Refuse what the permission profile does not allow at every mode check (28c5bb3)
- Record findings where checks run and link them to the failed step (5c5e6e4)
- Break spec 57 worker sessions into implementation tasks (eeab796)
- Plan spec 22 Task 8, recording findings where checks run (f587e7d)
- Block commands the permission profile does not allow (0956a9c)
- Apply a per-repository permission profile last and restrict-only (e829bba)
- Add the four permission profile capability keys (4e91737)
- Name the capability/preference partition of config keys (012b969)
- Plan spec 31 as a folder spec with a nine-task breakdown (3143cde)
- Stop config saves from dropping the command timeout and turn window (ea6301a)
- specs 57 preparation (5bed4b6)
- Persist findings with derived staleness and link them as plan evidence (bac63fc)
- specs update (9a5cb52)
- Keep the fake Slack test token from tripping GitHub push protection (5f05aff)
- Add v0.38.0 to the changelog (6cc309a)
v0.38.0
What's Changed since v0.37.0
- UX improvement (983e5b6)
- specs update (62b0435)
- Specify an off switch for automatic compaction in Settings (f1f4799)
- Turn browser diagnostics into findings (285a1f0)
- Plan spec 22 Task 6 against spec 12's typed browser report (73c02bc)
- Parse Biome diagnostics into findings (4d34efc)
- Parse cargo test failures into findings (d44c6b2)
- Plan spec 22 Task 4 against captured cargo test output (8bf27c6)
- Parse rustc and clippy diagnostics into findings (8c1314f)
- Plan spec 22 Task 3 against captured cargo output (3f0b101)
- Add finding parser dispatch with a generic fallback for unparsed failures (40e601e)
- Plan spec 22 Task 2 in full (3f35706)
- Add the redacting, bounding Finding type for spec 22 (5f52063)
- Close spec 12: document browser diagnostics and hand the typed report to spec 22 (ee254b0)
- Require approval for every remote browser diagnostic and show its real origin (0aa125c)
- Record spec 12's passing live run and its acceptance-criteria mapping (dded3da)
- Feed a batched tool round back as one assistant message with its reasoning (ae2afd1)
- Record the live re-run and the batched-round reasoning defect it found in spec 47 (c981f77)
- Stop classifying a completed stream that mentions error as a provider refusal (06f970e)
- Record spec 12's live browser run and the false refusal it found in spec 48 (34adca2)
- Show browser diagnostics as a card in the conversation (a3f76d1)
- Record, persist and stream browser diagnostics for the session (55dcffd)
- Redact GitHub and Slack tokens that directly follow a slash (58d0e7d)
- Render browser diagnostics for the model and stop counting page errors as tool failures (3f19fdb)
- Plan spec 22 as a folder spec with an eleven-task breakdown (4fb4f8a)
- Parse browser diagnostics into a typed, redactable report (633ec3b)
- Apply the session's mode to edit-request patch proposals (5c7cf8b)
- Close spec 20: document working modes and cover the AGENTS.md and allowlist criteria (ea9dd9e)
- Mark mode-refused command proposals rejected so violations are visible (0c10ad0)
- Plan spec 20 Task 10, the close-out task (0aa2247)
- Plan spec 20 Task 9 and find the audit-visibility gap for mode refusals (29dade3)
- Add the session mode control and its endpoint (837c95d)
- Stop command_allowlist from widening Plan and Review modes (1821bb4)
- Plan spec 20 Task 8, confirmed independent of Task 7 (e61ee84)
- Plan spec 20 Task 7 and find the allowlist mode-widening bug (826dc48)
- Refuse mode-forbidden actions in the orchestrator at all nine paths (7058a95)
- Plan spec 20 Task 6 and find the resume-path refusal gap (3f71ee6)
- Omit the command envelope from Ask's system prompt (f8016d0)
- Plan spec 20 Task 5 and record the DAMAIAN_EDIT_V1 scope gap (2ab0c93)
- Filter native tool list by session mode (c1ae8f7)
- Plan spec 20 Task 4 in full (1130269)
- Add session-mode persistence to SessionStore (3657830)
- Reorder spec 20 Tasks 3-4 and plan Task 3 in full (f6c58dd)
- Add MCP read-only capability to the working-modes permission matrix (df570b6)
- Expand spec 20 Task 2 into full TDD steps (c6c09af)
- Add SessionMode and the working-modes permission matrix (af20086)
- Split spec 20 into a folder and plan its task breakdown (37eaee9)
- Add v0.37.0 to the changelog (10efb54)
v0.37.0
What's Changed since v0.36.0
- Document prompt-cache accounting and close spec 49's slice 1 (7f4b2de)
- Add dormant explicit prompt-cache breakpoint switch (763badb)
- Add prompt cache prefix-stability guard tests (004d7f9)
- Add the cache_hit_rate eval-harness metric (bfca4c0)
- Surface prompt-cache hit rate and upper-bound cost in the shell and web UI (4078188)
- agents update (57a2274)
- Aggregate cached tokens per task with the rate's own denominator (0d2178a)
- Observe whether a provider reports a prompt-cache split (b8dd958)
- Observe whether a provider reports a prompt-cache split (c75eed2)
- Parse a provider's cache split into a subset of its input tokens (b648d5b)
- Record how many input tokens a provider served from cache (52b19d2)
- Add v0.36.0 to the changelog (0c6048f)
v0.36.0
What's Changed since v0.35.0
- Warn when a spec's dependency line contradicts that spec (8e41c7e)
- Record which specs are unblocked and what may be built in parallel (06a49bc)
- Close requirement 8's sequential, timing and batch-cancel gaps (a457669)
- Give the agent stoppable commands, batched reads, and a bounded continuation (7d5e0dc)
- Split the prompt cache spec into a folder and plan its accounting slice (7563980)
- Say that calls and rounds track one to one until batching lands (a2207fe)
- Count a run's tool rounds from the session log, and record the A/B that found it (746f4e4)
- Cover the new agent tools with an eval scenario and document their caps (9f25d0f)
- Give the agent ranged reads, listing, search and anchored edits (ebec372)
- Search file contents as a tool, capped and redacted (7cc1c8f)
- List repository paths as a tool rather than a shell command (7966330)
- Read a file by line range and bound what a read returns (5acc63e)
- Add session search and export to the web UI (4ccf53a)
- Serve session search and export over the shell (49e4758)
- Search sessions by text and export them, redacted with a count (11d88c7)
- Record a session's origin so server-mode sessions can be filtered (93b64ba)
- Record what the gate costs locally and how to scope it per task (5f34c08)
- Lift the tree walk out of the indexer so one walker serves both (fd4f33b)
- Design the agent working floor and plan its first slice (387e2f0)
- Close out the finished task plans and record how to do it as you go (78c916c)
- Add the v0.35.0 release row (611f45a)
- Specify changelog automation in the release gate spec (bf24a7f)
- Write the changelog row from the release pipeline (d700275)
v0.35.0
What's Changed since v0.34.0
- Document the process registry and the orphan sweep (f2619f3)
- Sweep orphaned processes at launch and on shutdown (fc21ab2)
- Record how to keep new tests off the filesystem watcher (45b65e8)
- Record the nextest runner change in the quality gate spec (28f4246)
- Format the curl transport test constructors (7ba54e9)
- Pin the nextest installer action to a current release (932f395)
- Run the test suite through cargo-nextest in the quality gate (46d14c3)
- test running time optimization (119d02a)
- Classify and retry provider refusals, with honest cost accounting (c12ccc2)
- Specify provider refusals and the fallback consent that follows (c9e6518)
- Kill registered processes on Ctrl-C through a self-pipe handler (b76c53f)
- Register PTY sessions so a crash does not leak a shell (99a99b8)
- Spawn shell commands in their own group and register them (44e3a13)
- Register curl model calls so a killed turn stops billing (c850000)
- Register MCP stdio servers so a crash cannot leak them (80cdef7)
- Kill a swept process group with SIGTERM before SIGKILL (f587c22)
- Record spawned children and refuse to kill a PID whose start time changed (e7c19e4)
- Read a process start time so a recycled PID is detectable (605ee01)
- Update rustls to 0.23.45 for RUSTSEC-2026-0285 (3bbdf51)
- specs update after revision (b272d04)
- Added changelog and specs for the new features. Extended specs with data from internal documentation (9665c9d)
- specs update (bbd9201)
- Re-present a plan review after a crash, and stop re-asking about a turn already resumed (6c17c7a)
- Measure plans in the harness and document what a plan means (83000d0)
- Show the plan, its steps, and what confirms each one (996876d)
- Review a plan before the first mutating step and carry it across a resume (f6accf6)
- Stop a turn before the call that would cross its token ceiling (347f189)
- Add a per-task token ceiling a repository may lower but not raise (cc6c08d)
- Tell a turn stopped for tokens from one stopped for rounds (4db0c37)
- Derive the task phase from step state so it cannot contradict it (5902bec)
- Give a non-trivial turn a plan and advance it as work lands (58307e2)
- Mint plan evidence from what a tool arm observed (99bae65)
- Persist a turn's plan in the session log and replay it back (4ebc971)
- Add the plan and evidence types a turn works through (7e1b018)
- Record what a tool reported, not just that it was dispatched (9afb6c3)
- Plan spec 21 against the code and give it the folder layout (599e82b)
- updated the baseline (c59e590)
- Close the live-tier tool-call defect and the metric it exposed (0991df9)
- Count a tool error as a call that failed, not one waiting on a human (92ba61b)
- Record the tool calls a run made, not the ones its script described (926049a)
- Pin the config rollback against a mistyped repository price key (eae01c2)
- Skip a repository config line Damaian cannot parse instead of failing the whole load (e4794d6)
- measuring costs against live (fb7d108)
- Record the measured cost figure and say why one rate may not fit a provider (26b59d4)
- Keep provider prices when config is saved, and stop one key erasing a built-in provider (aabf31b)
- Record what the first live-provider run measured and what it found broken (b653f2d)
- Offer the same tools in both tiers and record which set a run used (d7fd159)
- Stop reporting a markdown link as an unresolved file reference (2d4c7fa)
- Run the scenarios in the live tier instead of reporting an empty success (46de965)
- Report no data instead of zero when a tier measured nothing (d2764c0)
- Say where user config lives, next to the settings it holds (76c0e13)
- Document token accounting and close spec 19 (e9acd8a)
- Report real token figures in the eval harness (d994870)
- Let a user price their own tokens without shipping a price table (617d5a4)
- Show what each turn spent, marked when the figure is an estimate (2013af5)
- Count the model call a crash interrupted instead of losing its cost (5438e5a)
- Record what every model call spent, including retries and stopped turns (949ee15)
- Sum a task's token usage from append-only per-run events (326889c)
- Ask providers for token usage, and probe once for the ones that refuse (e474dbe)
- Read the provider's token figures when it reports them (b4756c3)
- Give every model run a token figure, estimated until the provider reports one (b083627)
- specs update (94c14e8)
- specs update (c784545)
v0.34.0
What's Changed since v0.33.0
- Tell the user what a crash interrupted and let them choose what happens to it (59bdf39)
- Note that the recovery sweep has no launch caller until spec 45 (8f44468)
- Document crash recovery in troubleshooting and close spec 17 (f127eae)
- Measure crash recovery end to end and unblock the harness resume scenario (b3d6270)
- Assert recovery for every task state and crash shape, and refuse to resume a recorded unknown outcome (00778df)
- Load sessions written before durable task state and classify their running tasks (39e96ed)
- Add recovery operations and refuse to resume an unknown outcome (5b212f6)
- Reattach a pending approval on restart or fail the task with a reason (49bb466)
- specs update (77980b6)
- Classify what a crash left behind and refuse to auto-resume unknown outcomes (6b52889)
- Record durable task state and action markers so a crash is classifiable (0ee32ec)
- specs update (118e464)
- screenshoot update (0c04ab5)
v0.33.0
What's Changed since v0.32.1
- Move the composer controls out of the input into a slim action row (c8384d5)
- Record the measured composer baseline and correct spec 44's effort levels (0e6d104)
- Add spec 44 for composer control density (fdb9854)
- Record the reviewed evaluation baseline (fe2dd44)
- Add the eval harness live tier, docs, and close spec 18 (08d9ec1)
- Add the eval harness deterministic tier and restrict top-level secrets/ (fdc8e30)
- wip (6e7539a)
- Replay reasoning_content on every assistant turn, not just native tool calls (04267b4)
- specs update (f8e7ca9)
- specs update (40b167e)
v0.32.1
What's Changed since v0.31.0
- Drop commit references from the spec progress tables (f48591a)
- Update spec 41 commit refs after the message rewrite (2ef0b18)
- Bring the README back in line with what ships (b45417a)
- Add danger variant to the specimen page and close spec 43 (3ddf911)
- Bring type onto the scale and slim the terminal chrome (50e084e)
- Size settings actions to content and mark the destructive ones (82d9519)
- Let the style guide page scroll (c4a4cb0)
- Add context disclosure specimen and record spec 42 (5ab84d2)
- Hide role labels and grow the composer from two rows (2739288)
- Show the active folder and session in the thread header (62aba26)
- Fold context files into the turn that read them (608c9b7)
- Promote the disclosure to a shared class (b6ce4ad)
- Add spec 42, conversation column density (eea344e)
- Pair the approval disclosure with its action row (dea9ac9)
- Size patch preview actions to their content (cd450c0)
- Restructure command approval and demote persistent grants (8e4b8e0)
- Add rendered specimen page for the UI style guide (31f4f29)
- Add button scale and scope the pre height minimum (2a733f1)
- Split spec 41 into a folder and correct its measured claims (8a69052)
- added style and specs for UI improvements (865161a)
- now user can rewind the state to the last checkpoint (762bdfd)
- specs update (9679f9d)
v0.31.0
What's Changed since v0.30.0
- added signing of binaries (1ee5200)
- small fix for typo and lint (2235b48)
- added the fix for potentialy malicious config being distributed over repository and overriding the user configuration (bd162f6)
- added specs for roadmap phases 0-6, plus a repository config trust boundary (f883f8c)
- added internal docs to gitignore (8514989)