feat: tracing dashboard redesign and robust project-memory tooling - #127
Merged
Conversation
ReadSkill now resolves a connected MCP server name (including plugin-style
aliases like designer-skill:designer-skill) to a bridge listing that server's
tools instead of "Skill not found", and ships a designer-skill stub that routes
frontend work to the MCP tools.
best_practices_always config option folds the full /best-practices engineering
guidance into the root session's system prompt at startup (default off), so the
guardrails apply to every new session without running the command.
StrReplaceFile returns a precise, actionable error when a multi-edit batch fails
schema validation (e.g. entries collapsed by a streaming glitch), naming the bad
entries and steering toward single-edit calls; valid edits are never partially
applied.
The default system prompt now requires absence claims ("no em-dashes", "no
leftover debug", "matches the source") to be backed by an actual zero-hit scan,
and to re-ask rather than act on a self-authored reading of a non-responsive
clarifying answer.
Also: /recap on|off toggle with grey autosuggest; recaps strip <system-reminder>
blocks; configured /login providers get distinct state styling; braille
background spinner and hanging-indent working tips; scratch files are cleaned up
on exception-path session exit. Removes leftover debug instrumentation from the
skill tool.
Unchanged context lines in file-edit diff snippets (Write, StrReplaceFile, and all diff cards) used a muted grey (tool_diff_context), which read as dimmed against the normal body text. Match the normal `text` token instead — terminal default foreground in dark, #213853 in light — so edited-file previews are easy to read. Added/removed lines keep their green/red styling, so changes still stand out.
Refresh the Statistics and Sessions surfaces of the agent tracing visualizer and add a new Usage page, keeping the existing neutral/zinc identity (no new brand accent) and the dependency-free SVG charts. - Add shared shadcn-style Card primitives (12px radius, subtle shadow) - Statistics: icon-tile metric cards with helper lines, titled chart/ tool/table cards, rounded bars, bordered hover-row project table, width-contained layout, and an empty state - Sessions: focus rings and a search hint in the toolbar, softer card radius with hover lift, folder-tile project group headers - New Usage page: summary cards, a GitHub-style activity heatmap keyed on daily turns (monochrome intensity ramp), and a turn-trend chart - Header: app icon tile, subtitle, accessible theme toggle; add the Usage tab as polished pills - Add a prefers-reduced-motion safety block to global CSS Behavior, data flow, and DOM/event contracts are unchanged.
Typing /report surfaced /report_error first because its "report" alias was an exact match, ranked above /reports (a command-name prefix match). Rank by match tier (name exact, name prefix, alias exact, alias prefix) then by command-name length, so the closest command name wins. /report now lists /reports first. Drop the now-unused _command_lookup and add regression tests.
Apply a soft-enterprise analytics treatment across the tracing visualizer and make the Usage visualizations feel intentional. - Introduce a single restrained blue accent (primary/ring tokens) for light and dark; charts and the heatmap now carry visual hierarchy - Shared premium MetricCard (rounded-2xl, icon tile, hover lift); used by Statistics and Usage - Usage: heatmap and a new Usage Insights panel sit side by side to use the available width; larger blue GitHub-style heatmap with a Turns/Sessions toggle and Less/More legend - New area trend chart with gradient fill, gridlines, axis labels, and a hover tooltip (dependency-free SVG) - Statistics: single accent tool bars with a neutral error badge instead of red segments; blue daily-usage series - Soft muted page background and consistent rounded-2xl cards Charts remain hand-rolled SVG (no recharts). Behavior and data flow are unchanged.
Align the web and vis frontends on the latest stable React (19.2.7, @types/react 19.2.17) on top of vite 8, and refresh both lockfiles. Both frontends type-check and build cleanly under vite 8 + react 19.2.7.
The chart used a fixed 600x140 viewBox with maxHeight, so preserveAspectRatio letterboxed it: the plot rendered ~600px wide and floated centered in a much wider card. Measure the container width with a ResizeObserver and render the SVG full width (taller, with gridlines and a sessions Y axis), matching the Usage trend chart.
…cation The project-memory budget check silently added a 3-char entry delimiter that the rejection message never disclosed, so a near-full store reported e.g. "2085/2200, entry (113) exceeds" — math that reads as satisfiable (2085+113<2200) but isn't. With no visibility into the true ceiling or what was stored, the agent could only blind-shrink the entry and loop until interrupted. - project_memory: delimiter-aware accounting; rejections now report exact free chars, the entry's real cost (content + separator), used/limit, and a compact inventory (index, size, preview) so the next remove/replace is guided. Add status() and capacity(); flag capacity failures via MemoryOpResult.full. - Memory tool: new read-only `list` action for mid-session introspection; on a full-store rejection, append a plain-language explanation (nothing lost, task continues, how to free space) to the user-facing tool card. - Raise limits MEMORY 2200->5000, USER 1375->2500 (within the 8 KB injection budget). - /memory: show per-store capacity and a "nearly full" guidance panel at >=85%. - memory.md: best-effort housekeeping guidance — don't loop on rejection.
Contributor
|
Warning Review limit reached
More reviews will be available in 44 minutes and 33 seconds. Learn how PR review limits work. Your organization has used up its prepaid credits, and credit purchases are no longer available. Enable the review add-on in the billing tab to keep reviews running — you're only billed for reviews past your plan's rate limits ($0.25/file). ⌛ How to resolve this issue?After more reviews become available, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available. Please see our Fair Usage Limits Policy for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Pro Plus Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (4)
📝 WalkthroughWalkthroughThis PR adds an always-on best-practices option, index-aware project memory accounting and gated memory writes, MCP skill-bridge lookup fallback, /recap on|off persistence and sanitization, stricter StrReplace multi-edit validation, shell/autocomplete and selector styling tweaks, and a new usage analytics frontend with charts and tests. ChangesMain Feature Integration
Estimated code review effort 🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Suggested labels
✨ Finishing Touches🧪 Generate unit tests (beta)
|
elkaix
changed the title
Add a structural detector that nudges the agent toward editing the authoritative project file when a memory write looks like a rule or value-assignment that belongs in a file (a recurring failure mode: a correction rephrased as a "preference" and stored in memory while the governing file stays stale). The guard never blocks — it only appends a one-line advisory on add/replace — so false positives cost a sentence. Also let replace/remove identify an entry by 0-based `index` (from `list`) as a deterministic alternative to `old_text` substring matching; out-of-range indices report the inventory so the retry is guided. Slash completer now surfaces the matched alias as the menu label (`/res` -> `/resume`) while keeping name matches ranked above alias-only matches. memory.md guidance updated for authoritative-files-first and the new index-based locator.
Regenerate the inline snapshots that pin auto-discovered state so they match the new code: - pyinstaller `hiddenimports` now includes the new `pythinker_code.tools.memory.routing_guard` module. - wire-handshake slash-command list: `/recap` usage gained `on|off` and the `skill:designer-skill` command was added. Make the recap slash-command test await type-safe: the registry types commands as `None | Awaitable[None]`, so guard with `isinstance(ret, Awaitable)` before awaiting (matching the existing shell-slash test helper) instead of awaiting the union directly, which pyright rejects.
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
Codecov Report❌ Patch coverage is 📢 Thoughts on this report? Let us know! |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 5
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
src/pythinker_code/ui/shell/prompt.py (1)
2198-2205:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winExact
/recapghost text is shadowed by the slash menu.When the buffer contains
/recapat line start,SlashCommandCompleterstill yields a/recapcompletion, Lines 2594-2606 auto-select it, and the Enter binding on Lines 2238-2242 accepts that menu item before submit. So the newexact_suggestionshook never drives the common/recap → /recap on|offflow. Either suppress slash completions once an exact suggestion exists, or have the Enter path preferbuffer.suggestionfirst.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/pythinker_code/ui/shell/prompt.py` around lines 2198 - 2205, The ghost `/recap` suggestion is being overridden by the slash menu; fix by preferring the exact auto-suggestion or suppressing slash completions: update the Enter/submit handler to check buffer.suggestion first and accept it if present (before calling the code that accepts a SlashCommandCompleter menu entry), and/or modify SlashCommandCompleter (or SlashCommandAutoSuggest usage) to skip emitting slash completions when exact_suggestions (the callable passed to SlashCommandAutoSuggest) returns a non-empty exact match. Target the Enter binding/submit routine and the SlashCommandAutoSuggest/SlashCommandCompleter logic so that buffer.suggestion is accepted first or slash completions are suppressed when an exact suggestion exists.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/pythinker_code/cli/__init__.py`:
- Around line 1206-1213: The cleanup block currently uses
contextlib.suppress(Exception) which silently hides import or runtime errors;
change it to an explicit try/except around the import and await
cleanup_session_scratch call, catch Exception as e, and log the failure at debug
level (include exc_info/stacktrace) while preserving best-effort semantics (do
not re-raise). Target the block that imports pythinker_code.scratchpad and calls
cleanup_session_scratch with _latest_created_session.work_dir,
session_id=_latest_created_session.id,
session_title=_latest_created_session.title, and ensure both import failures and
runtime failures are logged via the module logger (use existing logger variable)
so failures are visible without altering behavior.
In `@src/pythinker_code/soul/slash.py`:
- Around line 77-95: The code sets soul.runtime.config.tui.turn_recaps before
attempting persistence, so if load_config/save_config (functions load_config,
save_config) raise ConfigError/OSError the in-memory state remains changed;
modify the handler to capture the previous value (prev =
soul.runtime.config.tui.turn_recaps) before changing, then set
soul.runtime.config.tui.turn_recaps = enabled, attempt load_config/save_config,
and in the except block revert soul.runtime.config.tui.turn_recaps = prev and
still call wire_send(TextPart(...)) to report the failure; ensure the revert
happens for all persistence failures so runtime state matches the reported
outcome.
In `@src/pythinker_code/tools/memory/__init__.py`:
- Around line 90-93: The hardcoded full-store message always says “Project
memory … full” even when the store target is "user"; update the code that builds
the full-store guidance (the branch that references target and returns the
multi-line string) to choose target-specific wording: when target == "project"
keep the existing "Project memory…" text, when target == "user" change to "User
memory for this repo is full…" (or similar user-target phrasing), or build the
message from a small template that inserts the capitalized target name into the
sentence; ensure the conditional/template is used where the current multi-line
string is returned so the message accurately reflects the target variable.
In `@tests/core/test_session.py`:
- Around line 687-689: Rename the unused test parameter isolated_share_dir to
_isolated_share_dir in the two test function signatures that were modified (the
occurrences at the other nearby test as well) so Ruff ARG001 no longer flags it;
alternatively, add an explicit noqa comment to the parameter if you prefer to
keep the name. Ensure you update the parameter in both places where
isolated_share_dir was added (the two test definitions) and do not change any
other logic or uses (this parameter is only for fixture side-effects).
In `@tests/ui_and_conv/test_settings_recaps_slash.py`:
- Around line 62-80: The test test_recap_singular_on_persists_and_reloads
currently mocks shell_slash.save_config but doesn't assert it was invoked;
update the test to verify persistence by asserting the mocked save function was
called (e.g., assert shell_slash.save_config.called or assert
shell_slash.save_config.call_count == 1) after the Reload is raised and after
asserting config_for_save.tui.turn_recaps is True; reference the mocked symbol
shell_slash.save_config and the test function name when making the change.
---
Outside diff comments:
In `@src/pythinker_code/ui/shell/prompt.py`:
- Around line 2198-2205: The ghost `/recap` suggestion is being overridden by
the slash menu; fix by preferring the exact auto-suggestion or suppressing slash
completions: update the Enter/submit handler to check buffer.suggestion first
and accept it if present (before calling the code that accepts a
SlashCommandCompleter menu entry), and/or modify SlashCommandCompleter (or
SlashCommandAutoSuggest usage) to skip emitting slash completions when
exact_suggestions (the callable passed to SlashCommandAutoSuggest) returns a
non-empty exact match. Target the Enter binding/submit routine and the
SlashCommandAutoSuggest/SlashCommandCompleter logic so that buffer.suggestion is
accepted first or slash completions are suppressed when an exact suggestion
exists.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: 06fe983c-efb4-4e7b-99cd-457f58fb0d11
⛔ Files ignored due to path filters (2)
vis/package-lock.jsonis excluded by!**/package-lock.jsonweb/package-lock.jsonis excluded by!**/package-lock.json
📒 Files selected for processing (59)
CHANGELOG.mdsrc/pythinker_code/agents/default/system.mdsrc/pythinker_code/app.pysrc/pythinker_code/cli/__init__.pysrc/pythinker_code/config.pysrc/pythinker_code/project_memory.pysrc/pythinker_code/prompts/__init__.pysrc/pythinker_code/session_recap.pysrc/pythinker_code/skills/designer-skill/SKILL.mdsrc/pythinker_code/soul/slash.pysrc/pythinker_code/tools/file/replace.pysrc/pythinker_code/tools/memory/__init__.pysrc/pythinker_code/tools/memory/memory.mdsrc/pythinker_code/tools/memory/routing_guard.pysrc/pythinker_code/tools/skill/__init__.pysrc/pythinker_code/tools/skill/_mcp_bridge.pysrc/pythinker_code/tools/skill/description.mdsrc/pythinker_code/ui/shell/__init__.pysrc/pythinker_code/ui/shell/prompt.pysrc/pythinker_code/ui/shell/selector.pysrc/pythinker_code/ui/shell/selectors/oauth.pysrc/pythinker_code/ui/shell/slash.pysrc/pythinker_code/ui/shell/visualize/_live_view.pysrc/pythinker_code/ui/theme.pytests/core/test_best_practices_slash.pytests/core/test_builtin_authoring_skills.pytests/core/test_config.pytests/core/test_project_memory.pytests/core/test_session.pytests/core/test_slash_recap.pytests/test_session_recap.pytests/tools/test_mcp_skill_bridge.pytests/tools/test_memory_routing_guard.pytests/tools/test_memory_tool.pytests/tools/test_skill_tool.pytests/tools/test_str_replace_file.pytests/ui_and_conv/test_live_view_notifications.pytests/ui_and_conv/test_memory_slash.pytests/ui_and_conv/test_prompt_tips.pytests/ui_and_conv/test_selectors_simple.pytests/ui_and_conv/test_settings_recaps_slash.pytests/ui_and_conv/test_slash_completer.pytests/ui_and_conv/test_visualize_running_prompt.pytests/utils/test_pyinstaller_utils.pytests_e2e/test_wire_protocol.pyvis/package.jsonvis/src/App.tsxvis/src/components/metric-card.tsxvis/src/components/ui/card.tsxvis/src/features/sessions-explorer/explorer-toolbar.tsxvis/src/features/sessions-explorer/project-group.tsxvis/src/features/sessions-explorer/session-card.tsxvis/src/features/sessions-explorer/sessions-explorer.tsxvis/src/features/statistics/statistics-view.tsxvis/src/features/usage/usage-heatmap.tsxvis/src/features/usage/usage-trend-chart.tsxvis/src/features/usage/usage-view.tsxvis/src/index.cssweb/package.json
The react 19.2.7 / vite 8.0.16 bump left both lockfiles inconsistent:
they were missing the platform-specific optional @emnapi/* wasm-binding
deps (oxc-parser, oxc-resolver, rolldown). Because those are optional
per-OS deps, a partial `--package-lock-only` regen on one platform does
not capture the set CI's Linux runner needs, so `npm ci` kept failing
with EUSAGE ("package.json and package-lock.json are not in sync") and
broke `make build-web` / build-vis in the PyInstaller onefile jobs.
Regenerate both lockfiles from a clean install so they carry the full
cross-platform optional-dep tree. Direct deps are unchanged (react
19.2.7, vite 8.0.16, typescript 5.9.3); the churn is transitive/optional
ordering. Verified `npm ci` succeeds and `build-web`/`build-vis` produce
assets in both web/ and vis/.
elkaix
force-pushed
the
feat/enhancements-and-bug-fixes
branch
from
June 13, 2026 14:19
b1f32ca to
eaee384
Compare
- cli: replace silent contextlib.suppress(Exception) in the exception-path session cleanup with explicit try/except that logs at debug, so import or runtime failures during best-effort cleanup are traceable (CodeRabbit). - slash (/recap): revert the in-memory turn_recaps toggle when persistence fails, so runtime state matches the reported failure instead of silently diverging (CodeRabbit, major). - test_slash_recap: await the recap command via a type-honest cast instead of a bare `await ret` (clears the "statement has no effect" code-scanning flag). - test_session: silence ARG001 on the three exception-cleanup tests with @pytest.mark.usefixtures (renaming the param to _isolated_share_dir would break pytest fixture injection). - test_settings_recaps_slash: assert save_config is actually called in the singular `/recap on` alias test (CodeRabbit).
When an add/replace looks like a rule, a value/limit, or names a project file
(the existing routing_guard signals), gate it behind user confirmation before
writing — so the agent can't silently fill memory with corrections that belong
in an authoritative file. Plain durable facts (no signals) still write directly.
Uses the shared approval flow (Runtime.approval), so yolo/auto-approve and
"allow for session" are honored automatically:
- approved -> the entry is written;
- declined by the user -> not written; the agent is told to edit the governing
file instead, with the user's feedback;
- no user available (headless) -> skipped cleanly, reported as skipped (not an
error), so unattended runs never accumulate flagged memory.
This replaces the weaker post-write advisory nudge with a real gate. Also makes
the full-store guidance target-specific ("Project memory" vs "User memory")
instead of always saying "Project memory" (CodeRabbit).
Regenerating web/package-lock.json bumped ultracite within `^7.1.1` to 7.8.3, which restructured its package exports and dropped `ultracite/core` — web/biome.json extends `ultracite/core`, so `make check-web` (biome) failed with "Could not resolve ultracite/core". Pin ultracite to exactly 7.1.1 (the version main uses) until biome.json is migrated to newer ultracite. The cross-platform @emnapi lockfile fix is preserved; `npm ci`, biome, and the web build all pass.
Member
Author
|
Addressed the review feedback in 75f1927, f5eb04e, dcb1a19:
Not changed (with reason): the |
Member
Author
|
@coderabbitai review |
Contributor
✅ Action performedReview finished.
|
Contributor
There was a problem hiding this comment.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tests/core/test_slash_recap.py (1)
35-75: 🧹 Nitpick | 🔵 Trivial | ⚡ Quick winAdd a regression test for persistence failure rollback in
/recap on.Current coverage validates success and session-only behavior, but it doesn’t assert the rollback path when
load_config/save_configfails. Please add one negative-path test to lock in the Line 95-97 revert behavior.Proposed test shape
+from pythinker_code.config import ConfigError, get_default_config @@ +async def test_recap_on_reverts_runtime_when_save_fails( + runtime: Runtime, tmp_path: Path, monkeypatch: pytest.MonkeyPatch +) -> None: + config_path = (tmp_path / "config.toml").resolve() + runtime.config.source_file = config_path + runtime.config.tui.turn_recaps = False + soul = _make_soul(runtime, tmp_path) + sent: list[TextPart] = [] + + monkeypatch.setattr( + "pythinker_code.soul.slash.load_config", + Mock(side_effect=ConfigError("bad config")), + ) + monkeypatch.setattr("pythinker_code.soul.slash.wire_send", lambda msg: sent.append(msg)) + + await _run_recap(soul, "on") + + assert runtime.config.tui.turn_recaps is False + assert any("Failed to save recap setting" in msg.text for msg in sent)As per coding guidelines, “Reject or flag … tests covering only happy paths (C14).”
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/core/test_slash_recap.py` around lines 35 - 75, Add a negative-path test to tests/core/test_slash_recap.py that exercises rollback when persistence fails: create a test similar to test_recap_on_persists_and_updates_runtime but set runtime.config.source_file to a path and initial runtime.config.tui.turn_recaps = False, monkeypatch pythinker_code.soul.slash.load_config to return config_for_save and monkeypatch pythinker_code.soul.slash.save_config to raise an Exception (or mock that raises), monkeypatch wire_send to capture sent messages, call await _run_recap(soul, "on"), then assert that runtime.config.tui.turn_recaps was reverted back to False, that save_config was invoked (assert called_once), that config_for_save.tui.turn_recaps was not left True (or reverted), and that wire_send produced an error message (assert any("failed" in msg.text.lower() or "could not" in msg.text.lower() for msg in sent)).Source: Coding guidelines
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/pythinker_code/cli/__init__.py`:
- Around line 1206-1221: Remove the duplicate local import of logger inside the
try block: delete the line "from pythinker_code.utils.logging import logger" and
let the existing logger from the outer scope be used; keep the local import of
cleanup_session_scratch and the try/except unchanged so the await
cleanup_session_scratch(...) call and the logger.opt(exception=True).debug(...)
line reference the already-in-scope logger.
In `@src/pythinker_code/tools/memory/__init__.py`:
- Around line 116-118: The debug log in the memory approval gate currently logs
content preview (content[:60]) which may leak secrets/PII; update the logger
usage in the memory approval flow (the logger.bind(...).debug call) to remove
the raw content preview and log only structural metadata (e.g., target and
signals) or a constant redaction marker (e.g., "<REDACTED_PREVIEW>") instead;
ensure no other code in the same approval gate emits content before store-level
screening.
In `@web/package.json`:
- Line 84: The devDependency "ultracite" was pinned to an exact version
("7.1.1") which blocks automatic patch updates; revert it to a semver range by
changing the version string for "ultracite" in package.json back to "^7.1.1" (or
another appropriate caret/range) so dev tooling can receive patch/security fixes
automatically.
---
Outside diff comments:
In `@tests/core/test_slash_recap.py`:
- Around line 35-75: Add a negative-path test to tests/core/test_slash_recap.py
that exercises rollback when persistence fails: create a test similar to
test_recap_on_persists_and_updates_runtime but set runtime.config.source_file to
a path and initial runtime.config.tui.turn_recaps = False, monkeypatch
pythinker_code.soul.slash.load_config to return config_for_save and monkeypatch
pythinker_code.soul.slash.save_config to raise an Exception (or mock that
raises), monkeypatch wire_send to capture sent messages, call await
_run_recap(soul, "on"), then assert that runtime.config.tui.turn_recaps was
reverted back to False, that save_config was invoked (assert called_once), that
config_for_save.tui.turn_recaps was not left True (or reverted), and that
wire_send produced an error message (assert any("failed" in msg.text.lower() or
"could not" in msg.text.lower() for msg in sent)).
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: c703c9e0-8857-4114-bec3-9d9a2be85e86
⛔ Files ignored due to path filters (2)
vis/package-lock.jsonis excluded by!**/package-lock.jsonweb/package-lock.jsonis excluded by!**/package-lock.json
📒 Files selected for processing (9)
src/pythinker_code/cli/__init__.pysrc/pythinker_code/soul/slash.pysrc/pythinker_code/tools/memory/__init__.pysrc/pythinker_code/tools/memory/memory.mdtests/core/test_session.pytests/core/test_slash_recap.pytests/tools/test_memory_tool.pytests/ui_and_conv/test_settings_recaps_slash.pyweb/package.json
- memory gate: stop logging the raw content preview (content[:60]) in the routing-flagged debug log — a declined/blocked write could otherwise leak secrets/PII; log only target + signals (CodeRabbit, major). - cli: drop the redundant local `logger` import in the exception-path cleanup; the enclosing function already binds `logger` in scope (CodeRabbit). - docs/build: the react bump moved web/vis to Vite 8, but build_vis.py, web/ AGENTS.md, and architecture.md still said "Vite 7". Update the labels (the Node engine constraint ^20.19.0 || >=22.12.0 is unchanged in Vite 8).
This was referenced Jun 16, 2026
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Supersedes #126 (auto-closed when its head branch was renamed to
feat/enhancements-and-bug-fixes).Summary
This branch bundles the tracing-dashboard redesign with a robustness fix for the project-memory tool, plus supporting agent-UX and dependency work.
Project-memory capacity handling
A user hit a loop where the agent thrashed through 5 failed memory writes against a near-full store and had to be interrupted. Root cause: the budget check silently added a 3-char entry delimiter the error never disclosed, so
2085/2200, entry (113) exceedsread as satisfiable (2085+113 < 2200) but wasn't, and the agent had no way to see the true ceiling or what was stored.used/limit, and a compact inventory (index, size, preview) so the nextremove/replaceis guided, not guessed.listaction for mid-session introspection (status()/capacity()on the store)./memorynow shows per-store capacity and a "nearly full" panel at >=85%.Tracing dashboard redesign
/reportlists/reportsfirst).vis/webon vite 8.Verification
status/list, replace-overage, the/memorycapacity line, and the slash-completer ordering.vistypecheck + build green.Notes / tradeoffs
snapshot()already truncates gracefully and the header points the agent to the files. LeftINJECTION_BUDGET_BYTESat 8 KB to avoid inflating every prompt.Summary by CodeRabbit
New Features
Bug Fixes
UI/Style Improvements