core+web: agent-tagged log prefixes; one-shot capability hints; v0.9.50 - #196
Merged
Conversation
Log lines emitted inside a run now carry the running agent's name: CURRENT_AGENT (lovia.log_config) is set by RunLoop for the run's duration (updated on handoff, restored on exit), and the web run-source filter composes it with the run's source into [user/lovia], [user/memory-digest], [followups]-style prefixes — a side-run a chat triggers is no longer indistinguishable from the chat itself. The two static capability hints — 'context.window: unknown' and the returns_images 'not offering it' gate — now log once per process per model instead of on every run bootstrap: each helper side-run (follow-ups, titles, memory curation) repeated them after every chat exchange. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Pull request overview
This PR improves log attribution and reduces repetitive bootstrap diagnostics in lovia (core + lovia web) by (1) tagging logs with the currently-running agent name and (2) logging certain static capability/context-window hints only once per process per model. It also bumps the project version to 0.9.50.
Changes:
- Add
lovia.log_config.CURRENT_AGENT(contextvar) and set/reset it inRunLoop.stream()(and re-set on activation) so logs can be attributed to the active agent across handoffs and nested runs. - Compose web
RUN_SOURCE+CURRENT_AGENTinto a single%(run_source)sprefix (e.g.[user/lovia],[user/memory-digest],[followups]) for clearer console logs. - Introduce a loop-level
_log_oncememo to avoid repeating static “context.window unknown” / “tool … not offering it” hints across multiple runs; add tests for both behaviors.
Reviewed changes
Copilot reviewed 10 out of 11 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| uv.lock | Bump locked package version to 0.9.50. |
| pyproject.toml | Bump project version to 0.9.50. |
| lovia/init.py | Bump __version__ to 0.9.50. |
| lovia/log_config.py | Add CURRENT_AGENT contextvar for per-task agent tagging. |
| lovia/runtime/loop.py | Set/reset CURRENT_AGENT during runs; add _log_once for one-shot hints. |
| lovia/web/supervisor.py | Update run-source log filter to include agent tag in prefixes. |
| lovia/web/main.py | Update logging-format comment to reflect composed prefixes. |
| tests/workspace/test_workspace_agent.py | Add test asserting image-tool gating hint logs once per model. |
| tests/web/test_subagents.py | Extend tests to cover composed [source/agent] and agent-only tags. |
| tests/runtime/test_loop.py | Add tests for CURRENT_AGENT set/restore and nested-run shadowing. |
| tests/runtime/test_context_window_discovery.py | Add test asserting unknown context-window hint logs once per model. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Comment on lines
89
to
92
| def filter(self, record: logging.LogRecord) -> bool: | ||
| source = RUN_SOURCE.get() | ||
| record.run_source = f"[{source}] " if source else "" | ||
| tag = "/".join(p for p in (RUN_SOURCE.get(), CURRENT_AGENT.get()) if p) | ||
| record.run_source = f"[{tag}] " if tag else "" | ||
| return True |
Owner
Author
There was a problem hiding this comment.
Fixed in 3d062d7: the filter collapses control characters in the composed tag before formatting (isprintable() fast path, so the per-record cost stays a single check). Bracket characters are left alone — they can't fabricate lines, only look odd.
A free-form agent name containing a newline would split the one-line [source/agent] prefix and fabricate log lines. isprintable() keeps the per-record cost to a fast check; the collapse runs only on a pathological name. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Keep the cross-module contract notes (CURRENT_AGENT set by the loop, read by the web filter; the suppress(ValueError) edge) and drop restated rationale. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
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.
What
Follow-up to the log investigation: two changes that make
lovia webconsole logs attributable and quiet.Agent-tagged log prefixes
Every log line emitted inside a run now names the agent it belongs to. A new core contextvar
lovia.log_config.CURRENT_AGENTis set byRunLoop.stream()for the run's duration (re-set on handoff in_resolve_active, restored when the run ends; nested side-runs shadow and restore the host's tag). The web run-source filter composes it with the existing run source:[user][user/lovia][user][user/memory-digest][followups]/[titler][subagent:abc][subagent:abc/researcher]This is what made the original confusion possible: the memory digest's
context.window: unknownline rendered as plain[user]right after the chat'srun.done, reading as if the chat run had lost its configured window.One-shot capability hints
context.window: unknown for ...andtool 'view_image' ... not offering itare static facts about a model, not per-run events — but they re-fired on every bootstrap, including every follow-up/title/memory side-run. Both now log once per process per model (_log_oncememo in the loop), keeping the first occurrence's diagnostic value at INFO.Deliberately not done
Threading the configured context window into helper side-runs: one-shot runs have no history to compact (proactive budgeting is a no-op; reactive overflow rescue works without a window), and the correct wiring is not small — the followup/title model can be a different endpoint than the default profile, and the memory plugin is core so it cannot see web config. Assessment in the session; can be a follow-up if wanted.
Tests
CURRENT_AGENTset during a run, restored after; nested run shadows and restoresFull suite: 2310 passed, 41 skipped.
🤖 Generated with Claude Code