CodeLoupe 0.1.9

CodeLoupe 0.1.9

Changes since v0.1.8.

• CL-195 build the jars once for the macOS Intel bundle, split the Windows tests in three, uncompressed update test installers, settle child JVMs before the stopper probe; wait for the main method of a spinning child too; split the Linux tests in two, fix the flaky worker count, job takeover and stop races, fail a hung test by name; split the Windows tests in four, keep the update test installers compressed; describe the second round in docs/ci.md; build the three update test installers at the same time; build both update test bundles in one warm Gradle daemon; describe the parallel update test installers in docs/ci.md
• CL-226 stamp a session with its own login, not the shared login file; keep token components per minute and model, export them with metrics tokens; stamp plain CLI sessions from their login file again, export usage-limit answers
• CL-238 hand back the heap the serial collector grows: full collection after work, shrink without steps
• CL-249 give the AOT timing gate of the bundle smoke a 15 % margin over the dynamic archive
• CL-253 open SQLite files and start processes under a home past MAX_PATH
• CL-254 skip unpublished releases in the update check and follow the real daemon after a slow start
• CL-255 stamp the app version on the shipped plugin, unpin the repository copy, flag a stale plugin in settings

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.8

CodeLoupe 0.1.8

Changes since v0.1.7.

• CL-244 let a running job outlive its daemon on windows and be taken over by the next one; describe on the jobs wiki how a running job survives a restart on windows; check that no kill-on-close job object is made off windows
• CL-245 add slot capacity control, queue reorder and wait times to the jobs screen; document the jobs screen slot actions
• CL-247 parse large edits in batches on worker threads, read files ahead, open the merge-base index once per call
• CL-250 give the traced codeql build a 4 GB gradle heap
• CL-251 start the parse worker and the desktop app's daemon from AOT caches
• CL-252 override global-agent to 4.1.3 so the app no longer pulls sprintf-js through roarr

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.7

CodeLoupe 0.1.7

Changes since v0.1.6.

• CL-235 add brain risk map: per-file and module scores with the entries behind them; document brain_risk and round risk timestamps to seconds; retry the risk tests while the fixture code index builds and test the covered entry
• CL-239 drop taken, blocked, excluded and foreign tasks before cutting the dispatch_plan pool
• CL-240 remove only worktrees under the repository's roots when it names any
• CL-243 change slot capacity at runtime, reorder slot queues and keep queued jobs across a daemon restart; keep job wait waiting while a restarting daemon does not answer yet
• CL-246 overlap the git, tracker and disk waits of task_context and changes, copy merge-base facts from the base index
• CL-248 read the brains' task notes into the task detail and a tasks list per brain; show the brain notes in the task detail and a tasks view in the Brain screen; describe the task notes of the brain in the UI spec

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.6

CodeLoupe 0.1.6

Changes since v0.1.5.

• CL-240 read a worktree lock from its admin directory, remove links the walk cannot open; keep a process's memory out of the reconcile plan hash

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.4

CodeLoupe 0.1.4

Changes since v0.1.3.

• CL-234 add the brain index with code-linked ranking, anchors and the brain tools; read the brain topic view from show --json and test the brain index; anchor only declarations the text names, else their type; document the brain index; offer the brain tools only for a registered brain and keep feedback out of the index fingerprint
• CL-237 compare the brain repository by its real path so the symbol note shows behind a symlinked or short temp path

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.3

CodeLoupe 0.1.3

Changes since v0.1.2.

• CL-195 save the budgets config comment as UTF-8 so the post-edit sample test reads every source; use every core for test forks on CI, give Gradle 2 GB and skip the update test for Kotlin-only pushes; split the Windows tests over two runners; check that the slow test classes placed by hand in the shard script still exist; let the hook script test tolerate a script that exits without reading its input; keep two test forks on CI: four starved the timing-sensitive tests; balance the two Windows test shares and skip the update test for the test scripts of gradle/; split the macOS tests over two runners as well; keep the macOS tests on one runner: the macOS runners are scarce; let the hook script tests tolerate a script that exits without reading its input and the session start test accept the was line; run three test forks on the macOS runner; time the run list queries three times and judge the fastest in the ingest scale test; record the trial runs and the decisions in docs/ci.md; treat the AppImage that is being replaced as not yet installed in the update test; wait for the heavy lane to take the sync in the overlay test; add the later trial runs and the flaky tests that are left to docs/ci.md
• CL-209 document that the jobs screen lists daemon jobs only and that a wrapper fallback must say so
• CL-217 add post-edit check hook for syntax errors and unknown names
• CL-219 add deterministic module conventions card and conventions command
• CL-229 start daemons one at a time, create owner-only files exclusively and report an empty daemon answer
• CL-230 count the MCP server of the plugin as added and show the plugin version in the claude code card
• CL-231 attribute sessions to accounts by the desktop instance that started them
• CL-232 add the brain screen: workspace brains, topics, entries and search

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.2

CodeLoupe 0.1.2

Changes since v0.1.1.

• CL-197 record the planner file-coverage sentence and the suggester query= body in plan 8.4
• CL-198 count empty answers the agent accepted apart from gaps, and common words only for code files the answer named
• CL-199 record the cache, large-result and output-discipline analysis of the Terrio flow
• CL-202 record the planner and reviewer model and effort experiment (no variant passes)
• CL-203 add a knowledge store with code-anchored facts, search, stale detection and brain notation import; refuse an anchor path that leaves its checkout
• CL-204 record the ceiling table, go/no-go per item and the warm pool simulation in plan 8.6
• CL-205 add an optional batch tool for independent read-only calls and keep test temp dirs in one removable directory; record the batch tool A/B (rejected, median 0.97) in plan 8.10 and the agent run pages
• CL-206 count the final output of a streamed message instead of its first line
• CL-208 remove the temp files of a test run when it ends, sweep leftovers of killed runs, count the temp folder in CI; keep the test scratch directory in /tmp outside Windows so Unix socket paths stay short on macOS; sweep the directories of every older TestRepos prefix, not a fixed list; leave the temp folder entries of macOS system daemons out of the CI count
• CL-209 keep job records for at least 7 days, skip unreadable ones, log how many were found and test persistence across a forced end and a restart
• CL-210 accounts: token secret name, limit state from usage-limit answers in transcripts, cost of the session window; let the mock accounts test allow the tokenSecret field name
• CL-211 rewrite the agent runs part of the wiki for outside readers on its own page; split the agent runs page into a short summary and a details page, plain wording, takeaways and bottom line; correct the planner share and a keeping-versus-dropping sentence on the agent run pages
• CL-212 record the planner-only batch tool A/B (rejected, median 0.97) in plan 8.10a and the agent run pages
• CL-214 record the coder and tester cost-tail measurement and the governor decision
• CL-218 record the knowledge store A/B in the Terrio flow (summary agent, judge between stages, packs: not adopted) in plan 8.11a and the agent run pages
• CL-220 add metrics commands: which commands agents use and what they cost, as a CLI report and a Commands tab in the Runs screen; store the shell group, the input size and the Write answer at ingest, tell new from existing files by the answer, document the measurement
• CL-221 classify the build loops of coder runs and record the no-go in plan 8.12a
• CL-223 anchor facts that name a file or class without a directory and search the know store by anchor
• CL-224 remove default claude account flag, keep account management
• CL-226 store usage snapshots and the account of every transcript at ingest; stamp the login of every session at start and resolve accounts from it
• CL-228 run the plugin headers helper without a bare bash on windows

Other

• CL-206c record the recomputation of the A/B experiments with the corrected output rule
• FM-23 add phase 1 ceiling of the scripted tester to plan 8.13
• FM-23 add phases 2-4 of the scripted tester to plan 8.13

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.1

CodeLoupe 0.1.1

Changes since v0.1.0.

• CL-189 record the shell-search and compact-packet experiments for the Terrio coder and reviewer; note that the compact review packet is live in the Terrio workspace
• CL-190 add description, comments, callers and norms to the task_context pack; record the planner pack experiment and keep a failing optional part out of the pack; point the plan and wiki text at the coverage follow-up
• CL-191 let module take a directory prefix, say why an empty grep is empty, resolve header lines and id lists, stop counting range reads and tracker ids as gaps; update the parity golden for a header line, link the follow-up cards
• CL-192 forget what a worktree was read as when a session or subagent starts in it
• CL-193 neutral mock data, brand colours in the app and screenshots in the README
• CL-194 name the files earlier tasks changed along with the touched code and the callers of the members the task text names
• CL-195 run test classes in two forks and skip heavy CI steps for documentation-only pushes; record the first trial run in docs/ci.md; record the documentation-only trial run in docs/ci.md

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.0

CodeLoupe 0.1.0

Changes since v0.1.0-rc.1.

• CL-75 record the controlled starting-context runs of the CL-57, CL-58 and CL-59 changes
• CL-88 record the controlled measurements of waiting turns, workspace release and the GitNexus and IDEA retirement
• CL-92 record the controlled dispatcher, planner and issue-read benchmark in the plan
• CL-180 record why agents barely called CodeLoupe and the routing table that fixes it; remove merge markers from plan.md; record the live reviewer and planner edits and the planner confirmation run
• CL-182 record the controlled re-run of the suggester and planner bodies in the plan; show the task state and epic in dispatch_plan window and waiting rows; record the dispatch_plan state-row measurement of the suggester in the plan; correct the task-call figures in the state-row measurement
• CL-184 trim the descriptions of the CodeLoupe tools the roles carry: schemas 0.65 -> about 0.48 percentage points of agent cost
• CL-185 record the job start --wait experiment: no waiting turns, one model call per long command

Other

• docs add the Agent runs section to the Benchmarks wiki page and link it from the README
• docs add the job start --wait and tool-schema trim results to the Agent runs section

Verify the files

• Checksums: sha256sum -c SHA256SUMS.txt in the folder with the downloaded files.
• Provenance (built by this repository's release workflow from this tag): gh attestation verify <file> --repo Terrio-cz/CodeLoupe.

CodeLoupe 0.1.0-rc.1

CodeLoupe 0.1.0-rc.1

First release.

• CL-7 record the decision for a standalone electron app
• CL-9 add kotlin index, daemon and find/outline/symbol tools
• CL-10 copy the facts of a file from a store that holds the same content instead of parsing it again
• CL-11 add the Java adapter: facts of Java files through the compiler's Java PSI; index Java sources: statics, Java collections, fixtures, tests for every query, docs and measurements
• CL-13 add usages: references labelled exact or candidate, a superset of rg -w; fix review findings: scopes, receivers, promotion and caches; fix second review round: index format, library supertypes, invocable bindings, lambda parameter receivers; fix third review round: receivers through nested lambdas and of receiver names
• CL-14 add calls and hierarchy: caller and callee trees, supertypes, subtypes and overrides
• CL-15 CL-18 add the grep and context tools; document grep, context and the status budgets
• CL-16 add worktree overlays and lazy base sync; fix review findings: shared waits, sync supersede, name-based gc, bounded scans, sparse files; fix second review round: unreadable dirs, sparse narrowing, large file hashes, moved base
• CL-17 add the changes tool: declarations changed against the merge-base with callers and tests; fix review findings: per-call missing files, bounded output and memory, overload redirects, kdoc, cli; fix second review round: absent blobs, callers budget, overload fqn, type body diffs; test the callers limits, foreign same-named functions and absent git objects
• CL-19 add the ranked repository map to outline
• CL-20 add the usages golden test: 40 Terrio symbols against a manually verified oracle; extend the golden set with callable references, smart casts, library extensions and this in apply
• CL-21 add codeloupe metrics collect and compare, read from Claude Code transcripts; document codeloupe metrics, the gap report and the scaffold no-go
• CL-22 add codeloupe metrics gaps, the weekly gap report
• CL-23 record the controlled planner and reviewer benchmark in the plan; record the reviewer benchmark rerun after the CL-179 body fix
• CL-24 add latency percentiles, budgets and resource history to status; leave the test JVM's rss out of the daemon test's budget check
• CL-25 add the load test, copy the base outside the repository lock, record the measurements
• CL-26 mirror tracker projects locally and answer issue and task queries; fix review findings: paged catch-up re-check, caller-only read memory, history on watermark move; close a mirror connection that fails to set up
• CL-28 add slim tracker write: update tool answers changed fields in one line
• CL-31 add starting-context audit
• CL-32 add output compressors for git, gradle and test runners; add run: summaries of command output with a job handle, and doc reads the handle; shorten the run header, group status entries of one kind; remove the throwaway compression dump test that fails on CI: it reads files from a local path and writes to TEMP
• CL-35 add codeloupe metrics boilerplate to measure the skeleton share of new files
• CL-36 add the lazy member merge prototype and the study that measures it; record the lazy edit comparison and the no-go decision
• CL-37 add the edit tool: structural writes, rename, write policy and gate, with the round-trip, fuzz and rename harness; keep the write tests out of .git, where git's background maintenance changes files while they walk
• CL-38 add desktop ui spec with wireframes, tokens and read-only api contract; address spec review: contrast tokens, cl-54 flows, ipc hardening, paged steps, nav and events; address second spec review round; point run budget highlight at overbudget flag; keep the edit menu on macos
• CL-39 add the read-only UI API in the daemon: nav, overview, worktrees, tasks, index, gaps, environment, settings, events; record the worktree of a call as git spells its path
• CL-40 add the screen registry, overview latency and resource charts with budget warnings, and the weekly gap report; add the runs screen over the transcript ingest, drop the gap report refresh the ingest made redundant, warm the worktree list, widen the screenshot run; describe the screens in the readmes
• CL-41 list the costliest runs of the branch's task in its drawer, with links to the run and back to the task
• CL-43 add electron desktop app with tray, daemon management and mock api; drop agent runs from the app and spec, add call telemetry, tests and readme; address app review: settle aborted responses, stop marker, notifications, a11y and markdown hardening; address second app review: opaque task cursors, marker timing, path lookup, accounts contract; capture painted frames and report private memory; use the electron app id for toasts in development runs
• CL-45 add claude code plugin and connect from the desktop app
• CL-50 add the secret store, the env tool and CLI, env run, and masking of stored values
• CL-51 document the secret store, rename the token file, test every surface; fix the tool list order in the daemon test
• CL-52 import existing env variables into the store from Claude folders
• CL-54 add, rotate, delete and import environment keys from the desktop app
• CL-55 audit secret reads and changes by consumer, flag keys past the rotation age
• CL-56 record the decision to port codeloupe to kotlin; record the parser spike and choose kotlin psi; port codeloupe to kotlin/jvm with parity to the prototype; fix review findings: daemon cwd, robust builds, index format, ordering, api errors; fix second review round: ref lookups, build timeouts, log content; remove the node prototype; point the desktop app at the kotlin cli
• CL-59 record the measured main-session payload trim
• CL-62 ingest agent transcripts into the daemon: runs, steps, budgets, gaps and events for the desktop app
• CL-63 wip accounts; manage Claude and YouTrack accounts: usage per account, overview filter, tokens only in the store; look for an account's transcript folder under the real path of a working directory too
• CL-64 start the CLI from a class-data archive with a plain HTTP client
• CL-66 add workspace registry with orphan detection, GET /workspaces and workspaces command
• CL-67 label docker resources by workspace with ws up, run, build, volume and report unowned ones
• CL-68 add reconciler that cleans released workspaces with retries, dry run, protect rules and journal
• CL-69 add ws release: mark a workspace released at once, the reconciler cleans it in the background
• CL-70 add port registry per workspace with conflict reports and ws ports
• CL-71 wip; track processes by workspace, report memory per workspace and stop idle build daemons of released workspaces; read the whole command line and the cpu time on Linux and macOS, resolve paths that no longer exist, settle test processes before probing them; resolve the existing part of a path that is gone before comparing it with a workspace
• CL-72 add the workspaces screen with the registry states, resources, ports, disk and memory, release and confirmed cleanup
• **[CL-78](https://github.com/Terrio-cz/CodeLoupe/commits?q=CL...