Match rotation semantics to prior behavior by fully resetting the dedupe map
once the size limit is exceeded, and add focused tests for duplicate detection
and boundary rotation behavior.
Centralize dedupe map access behind a mutex-safe helper and use it in both
WeCom bot and WeCom app channels to eliminate concurrent map access races while
preserving current dedupe behavior.
- classifier.go: s/honour/honor/ (American English per misspell)
- router.go: break SelectModel signature across lines (golines)
- router_test.go: break long Message literal (golines)
- router_test.go: replace CJK string literal with rune slice so
gosmopolitan does not flag the source file; behaviour is identical
instance.go:
- Add Router *routing.Router and LightCandidates []FallbackCandidate
to AgentInstance.
- At agent creation, when routing.enabled and light_model resolves
successfully in model_list, pre-build the Router and resolve the
light model candidates once. If the light model isn't in model_list,
log a warning and disable routing for that agent gracefully.
loop.go:
- Add selectCandidates(agent, userMsg, history) helper.
It calls Router.SelectModel and returns either agent.Candidates /
agent.Model (primary tier) or agent.LightCandidates / light_model
(light tier). Returns primary unchanged when routing is disabled.
- In runLLMIteration, resolve (activeCandidates, activeModel) once
before entering the tool-iteration loop. The model tier is sticky
for the entire turn so a multi-step tool chain doesn't switch
models mid-way.
- Replace hard-coded agent.Candidates / agent.Model references in
callLLM and the debug log with the resolved active values.
The fallback chain and retry logic are untouched. When light_model
returns an error the fallback chain handles escalation normally.
Add three new files to pkg/routing/:
features.go — ExtractFeatures(msg, history) → Features
Computes five structural dimensions with zero keyword matching:
- TokenEstimate: rune_count/3 (CJK-safe token proxy)
- CodeBlockCount: ``` pairs in the message
- RecentToolCalls: tool call count in the last 6 history entries
- ConversationDepth: total messages in session
- HasAttachments: data URIs or media file extensions
classifier.go — Classifier interface + RuleClassifier
RuleClassifier uses a weighted sum that is capped at 1.0:
code block → +0.40 (triggers heavy model alone at 0.35 threshold)
token > 200 → +0.35 (triggers heavy model alone)
tool calls > 3 → +0.25
token 50-200 → +0.15
conversation depth > 10 → +0.10
attachment → 1.00 (hard gate, always heavy)
router.go — Router wraps config + Classifier
Router.SelectModel(msg, history, primaryModel) returns either the
configured light_model or the primary model depending on whether
the complexity score clears the threshold. Threshold defaults to
0.35 when zero/negative to prevent misconfiguration.
router_test.go — 34 tests covering all branches and edge cases
Introduce RoutingConfig with three fields:
- enabled: activates per-turn model routing
- light_model: references a model_name in model_list
- threshold: complexity score cutoff in [0,1]
When routing.enabled is true and the incoming message scores below
threshold, the agent switches to light_model for that turn. Absent or
disabled config leaves existing behaviour completely unchanged.
Example:
"agents": {
"defaults": {
"model": "claude-sonnet-4-6",
"routing": {
"enabled": true,
"light_model": "gemini-flash",
"threshold": 0.35
}
}
}
Without this function, media:// refs stored by MediaStore are passed
directly to the LLM API, which rejects them as invalid URLs.
resolveMediaRefs() runs after BuildMessages() and before the LLM call,
converting each media:// ref to a data:image/...;base64,... URL that
vision-capable models can process.
Also adds mimeFromExtension() helper for MIME type inference from
file extensions when ContentType metadata is not available.
- Introduced WeCom AIBot channel configuration in config.go with relevant fields.
- Implemented WeCom AIBot channel factory registration in init.go.
- Created unit tests for WeCom AIBot channel functionalities including initialization, start/stop behavior, webhook path handling, message encryption/decryption, and signature generation.
- Set default values for WeCom AIBot configuration in defaults.go.
The serializeMessages() function was not preserving the reasoning_content
field when serializing messages for vision API calls. This caused the
TestProviderChat_PreservesReasoningContentInHistory test to fail.
This fix ensures reasoning_content is included in both text-only messages
and vision messages with media attachments.
Co-authored-by: Zachary Guerrero <zack.grrr@gmail.com>
Add Media field to processOptions, pass msg.Media from inbound
messages through to BuildMessages and serializeMessages so
vision-capable LLMs receive image_url content parts.
Based on work by @as3k in sipeed/picoclaw#555.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add Media []string field to Message struct for image/media URLs
- Implement serializeMessages() to format messages with image_url content parts
- Enables OpenAI-compatible vision APIs to receive image attachments
* fix(tools): allow /dev/null redirection and add read/write sandbox split
- Remove deny pattern that incorrectly blocked redirects to /dev/null
- Expand block device write pattern to cover nvme, mmcblk, vd, xvd,
hd, loop, dm-, md, sr and nbd in addition to sd
- Add safe path whitelist for kernel pseudo-devices so workspace path
check does not reject /dev/null, /dev/zero, /dev/random, /dev/urandom,
/dev/stdin, /dev/stdout and /dev/stderr
- Add allow_read_outside_workspace config option (default true) so file
read and list tools are unrestricted while write tools stay sandboxed
Closes https://github.com/sipeed/picoclaw/issues/964
Closes https://github.com/sipeed/picoclaw/issues/965
Signed-off-by: Huang Rui <vowstar@gmail.com>
* feat(tools): add configurable allow patterns and path whitelists
- Add custom_allow_patterns to exec config so users can exempt specific
commands from deny pattern checks
- Add allow_read_paths and allow_write_paths regex lists to tools config
for whitelisting specific paths outside the workspace
- Introduce whitelistFs that wraps sandboxFs and falls through to hostFs
for paths matching whitelist patterns
- Use variadic constructor signatures to keep backward compatibility
Suggested-by: lxowalle
Signed-off-by: Huang Rui <vowstar@gmail.com>
---------
Signed-off-by: Huang Rui <vowstar@gmail.com>
* fix: return fetched content to LLM in web_fetch tool
WebFetchTool.Execute was setting ForLLM to a summary string
("Fetched N bytes from URL ...") instead of the actual extracted
text. This meant the LLM never saw the page content and could not
answer questions based on fetched web pages.
Return the extracted text in ForLLM so the model can use it.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: put full JSON result in ForLLM, summary in ForUser
Accept suggestion from afjcjsbx: the LLM should receive the full JSON
result (including extracted text) while the user sees a short summary.
Update tests to match the new field assignment.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat(config): Add support for env var configuration
This commit introduces support for two environment variables,
allowing users to override the default paths for picoclaw's home
directory and configuration file.
- `PICOCLAW_CONFIG`: Directly specifies the path to the `config.json` file.
This is initialised first, takes precedence over the hardcoded path, and is ideal
for containerized deployments or custom config management.
- `PICOCLAW_HOME`: Overrides the root directory for all picoclaw data, (except the config)
(e.g., `~/.picoclaw`). This is useful for portable installations or placing
data in non-standard locations.
This change provides greater flexibility for running picoclaw in various environments without
being tied to the default home directory structure.
* `README.md` updated explain PICOCLAW_CONFIG and PICOCLAW_HOME
* docs: translate environment variables section to multiple languages
---------
Co-authored-by: picoclaw <picoclaw@sipeed.com>
- Remove Feishu from webhook channel list in README.md and README.zh.md;
add clarifying note that Feishu uses WebSocket/SDK mode instead
- Replace Chinese text in README.vi.md header with Vietnamese equivalent
- Translate mixed-language WeCom note in README.vi.md to full Vietnamese
- Mark webhook_path as optional (否) in docs/channels/line/README.zh.md
- Remove incorrect yaml struct tags from new-channel example in
pkg/channels/README.md and README.zh.md (config uses json tags only)
- Fix multi-mode initChannel example to use whatsapp/whatsapp_native
(matching the "WhatsApp Bridge vs Native" comment) instead of matrix
- Correct ReasoningChannelID description: list the 12 channels that
have the field and note that PicoConfig does not expose it
* fix(pkg/providers):do regex precompile insteadd on the fly
* fix(providers): replace HTTP-specific regex with standalone status code matcher
The precompiled HTTP regex used uppercase "HTTP" which never matched
because ClassifyError lowercases the input. Replace it with a
case-insensitive word-boundary pattern that matches any standalone
3-digit status code (300-599), which also subsumes the HTTP/x.x case.
Add test case for standalone status code extraction.
* fix(providers): restore http regex and add standalone status code matcher
Restore the http-prefixed regex (without unnecessary (?i) flag since
input is already lowercased by ClassifyError) as a mid-priority pattern
to reduce false positives. Add a standalone word-boundary matcher as a
fallback for bare status codes like "429". Fix test to use lowercased
input matching the actual calling convention.
* perf(tools): move path regex compilation from per-call to package init
The path regex in guardCommand was compiled on every call. Hoist it
to a package-level var (absolutePathPattern) alongside defaultDenyPatterns
in a single var block, so it is compiled once at init time.
* style(tools): move inline comment to fix golines formatting error
- Fix ignored error from SendAndWait call
- Improve error message for unimplemented stdio mode with helpful guidance
- Add TODO comment with reference link for future stdio implementation
The openaiMessage struct and stripSystemParts() were not carrying over
the ReasoningContent field when serializing conversation history for
API requests. This caused thinking models (e.g. kimi-k2.5) to receive
incomplete assistant messages on subsequent turns, resulting in 400
errors from the Moonshot API.
Add the ReasoningContent field to openaiMessage and copy it in
stripSystemParts(). Also add a test to verify reasoning_content is
preserved when sending conversation history.
Fixes#588
Related: #876
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace manual temp+rename in writeMeta and rewriteJSONL with the
project's standard fileutil.WriteFileAtomic. This adds fsync before
rename, which is important for flash storage on embedded devices
where power loss can leave zero-length files after an unsynced rename.
- Log a warning when readMessages skips a corrupt line, so operators
can see that data was lost after a crash instead of silently dropping it.
- Document the lossy sanitizeKey mapping (telegram:123 → telegram_123)
as an intentional tradeoff.
* fix(tools): close resp.Body on retry cancel and cache http.Client instances
Fix resp.Body leak in DoRequestWithRetry where req.Body (request) was
incorrectly closed instead of resp.Body (response) on context cancel.
Cache http.Client on web search/fetch provider structs and channel
adapters (WeCom, LINE) to avoid per-call allocation overhead.
* fix(channels): preserve original http client timeouts for LINE and WeCom
Split LINE single 60s client into infoClient (10s) for bot info lookups
and apiClient (30s) for messaging API calls. Lower WeCom cached client
base timeout from 60s to 30s (matching uploadMedia), and ensure it is
always >= the configured ReplyTimeout so the per-request context
deadline remains the effective limit.
* refactor(tools): extract timeout consts and deduplicate WebFetchTool constructors
Address PR review feedback from xiaket:
- Define searchTimeout, perplexityTimeout, fetchTimeout, defaultMaxChars,
and maxRedirects as package-level consts instead of magic numbers.
- Remove misleading "No proxy" comment in NewWebFetchTool.
- Deduplicate NewWebFetchTool by delegating to NewWebFetchToolWithProxy.
* test(utils): add context cancellation test for DoRequestWithRetry
Verify that resp.Body is properly closed when the context is canceled
during retry sleep, covering the C8 resp.Body leak fix.
* fix(utils): close resp in test to satisfy bodyclose linter
* fix(utils): eliminate flakiness in context cancellation retry test
Synchronize cancellation using an onRoundTrip callback from the
transport wrapper instead of a timing-based context timeout. This
ensures the first client.Do completes before cancel fires, so
cancellation always hits during sleepWithCtx.
- MCPTool.Name(): append FNV-32a hash of original (unsanitized) server+tool
names whenever sanitization is lossy or total length exceeds 64 chars,
ensuring names that differ only in disallowed characters remain distinct
- ToolRegistry.Register(): emit warn log when a tool registration overwrites
an existing tool with the same name, making collisions observable
- scripts/test-docker-mcp.sh: switch shebang from #/bin/bash /Users/yuchou/Work/klook-calendar/klook-google-cal-sync/src/googlecalconversrv/bin/start.sh to # for portability on minimal distros and Nix environments
- Avoid logging sensitive cfg.Args in ConnectServer; log args_count instead
- Sanitize server/tool name components in MCPTool.Name() to ensure valid
identifiers for downstream providers (lowercase, [a-z0-9_-] only)
- Add slack as 5th MCP server example in config.example.json
- Move Dockerfile.full and docker-compose.full.yml into docker/ directory
for consistency with existing docker/Dockerfile and docker/docker-compose.yml
- Fix all Makefile docker-* targets to reference correct compose file paths
- Fix docker/docker-compose.full.yml build context (.. ) and volume paths
- Fix scripts/test-docker-mcp.sh compose file path and replace cowsay test
with actual @modelcontextprotocol/server-filesystem MCP server test
Cancel the constructor-created context before overwriting in Start()
to prevent the original cancel function from becoming unreachable.
Move len(processedMsgs) check inside the write lock to eliminate a
data race, and re-insert the current msgID after map reset to prevent
duplicate processing of the in-flight message.
Applies to both WeComBotChannel and WeComAppChannel.
The ctx field was only set in Start(), so tests calling handleMessageCallback
without Start() caused a nil pointer dereference in MessageBus.PublishInbound.
The HTTP request context is canceled as soon as the handler returns the
response, causing PublishInbound to fail with "context canceled" when
processMessage runs asynchronously in a goroutine. Use the channel's
long-lived context (c.ctx) instead.
A standalone web-based tool for managing picoclaw configuration, OAuth
authentication providers, and gateway process lifecycle. Features include
a sidebar layout with i18n (en/zh) and theme support, real-time gateway
log streaming, startup prerequisites checks, and Windows icon embedding.
Co-authored-by: wj-xiao <meetwenjie@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
- WhatsApp Start(): use deferred cleanup to nil out c.client/c.container
and disconnect/close resources on any error after struct fields are
assigned, preventing stale references and double-close in Stop()
- handleReasoning: treat bus.ErrBusClosed as an expected condition
(DEBUG level) alongside context timeout/cancel, avoiding WARN noise
during normal shutdown
- WhatsApp Send(): detect unpaired state (Store.ID == nil) and return
ErrTemporary instead of attempting to send while QR login is pending
- handleReasoning: check the returned error type (DeadlineExceeded /
Canceled) instead of ctx.Err() to decide log level, so pubCtx
timeouts on a full bus are correctly classified as expected
- Test: fill bus with a short-timeout loop instead of hardcoding the
buffer size (64), making the test resilient to buffer size changes
Move the stopping check and wg.Add(1) inside reconnectMu in
eventHandler, and set the stopping flag under the same lock in Stop().
This makes the two operations atomic with respect to each other,
preventing the race where:
1. eventHandler checks stopping (false)
2. Stop() sets stopping=true and enters wg.Wait() (wg is 0)
3. eventHandler calls wg.Add(1) → panic or goroutine leak
- Use c.runCtx for GetQRChannel so the QR producer is canceled on Stop()
- Add atomic stopping guard to prevent wg.Add/wg.Wait race in eventHandler
- Make Stop() context-aware: disconnect client before waiting, respect ctx deadline
- Reduce reasoning publish log noise: use debug level for expected ctx errors
- Add test for handleReasoning when outbound bus is full (timeout path)
Add a 5-second timeout to handleReasoning's PublishOutbound call so
fire-and-forget goroutines do not block indefinitely when the outbound
bus channel is full. Reasoning output is best-effort; on timeout the
publish is abandoned with a warning log instead of holding the
goroutine alive.
Fixes goroutine leak introduced in #802.
- Move runCtx/runCancel creation before event handler registration and
QR loop so Stop() can cancel at any point during startup
- Replace blocking QR event loop in Start() with a background goroutine
that selects on runCtx.Done(), preventing Start() from hanging
indefinitely when waiting for QR scan
- Track all background goroutines (QR handler, reconnect) with
sync.WaitGroup; Stop() waits for them to finish before releasing
client/container resources
- Cancel runCtx on error paths in Start() to avoid leaked contexts
Fixes resource leak introduced in #655.
Define PlaceholderCapable, TypingCapable, and ReactionCapable interfaces
and have BaseChannel.HandleMessage auto-detect and trigger all three as
independent pipelines on inbound messages. This replaces the scattered
manual orchestration code in each channel's handleMessage with a single
unified dispatch in the framework layer.
Changes:
- Add PlaceholderCapable interface to interfaces.go
- Add ReactionCapable + RecordReactionUndo to interfaces.go
- BaseChannel.HandleMessage auto-triggers Typing → Reaction → Placeholder
- Manager gains reactionUndos sync.Map with TTL janitor cleanup
- Telegram: extract SendPlaceholder from manual code, add StartTyping
- Discord: add SendPlaceholder + StartTyping
- Pico: add SendPlaceholder (uses Pico Protocol message.create)
- Slack: extract ReactToMessage from manual code
- OneBot: extract ReactToMessage, remove leaked pendingEmojiMsg sync.Map
- LINE: move group-chat guard into StartTyping, remove manual orchestration
- Config: add Placeholder to PicoConfig; remove from Slack/LINE/OneBot
(no MessageEditor, so placeholder config was dead code)
Port changes that were applied to the old pkg/channels/*.go files on main
to their new locations in channel subpackages:
- telegram: precompile regex, var transcribedText, GetModelName()
- discord: var transcribedText declaration
- onebot: resp.Body.Close(), "canceled" spelling, remove empty line
- slack: named return values in parseSlackChatID
- wecom: remove sendMarkdownMessage dead code
- whatsapp: resp.Body.Close() after Dial
- gateway/helpers: remove unused errors import
Prevents potential backpressure under load when multiple channels
publish concurrently in gateway mode, where SDK callbacks blocking
on a full buffer can cause message loss or timeouts.
- Replace stdlib log.Printf with logger.InfoCF/WarnCF for consistency
with the rest of the codebase (addresses @nikolasdehor review point #3)
- ReleaseAll: clean refToScope/refs mappings even if refs entry is missing
- CleanExpired: guard refToScope lookup before scope cleanup
- Add TestReleaseAllCleansMappingsIfRefsMissing for robustness
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Address review feedback from @alexhoshina and Codex:
- Log sendLoading errors in ticker goroutine instead of discarding
- Only start typing indicator when PlaceholderRecorder is available
to avoid wasted API calls and unnecessary goroutine creation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(skills): add retry mechanism for HTTP requests
Implement a retry mechanism with exponential backoff for HTTP requests in the skill installer. This improves reliability when fetching skills from GitHub by automatically retrying failed requests up to 3 times.
Add comprehensive tests to verify retry behavior under different scenarios including success on different attempts and proper delay between retries.
* fix: improve http request retry logic with status code checks
Add shouldRetry helper function to determine retryable status codes.
Close response body between retry attempts and break early for non-retryable status codes.
* refactor: remove unused BuiltinSkill struct
The struct was not being used anywhere in the codebase, so it's safe to remove it to reduce clutter and improve maintainability.
* refactor(http): move retry logic to utils package
Extract HTTP retry functionality from skills package to utils for better reusability
Add context-aware sleep function and comprehensive tests
* refactor(http): extract retry delay unit to variable
Extract hardcoded retry delay unit to a variable for better testability and flexibility. Update tests to use milliseconds for faster execution while maintaining the same behavior.
* test(http_retry): remove t.Parallel from test cases
* test(http_retry): remove redundant test cases for retry success
The removed test cases for success on second and third attempts were redundant since the retry logic is already covered by other tests. This simplifies the test suite while maintaining coverage.
Change the default value of session.dm_scope from "main" to
"per-channel-peer" to provide better conversation isolation by
default. This prevents context leakage between different users
and channels.
MigrateFromJSON previously called AddFullMessage in a loop, then
renamed the .json file to .json.migrated. If the process crashed
after appending some messages but before the rename, a retry would
re-read the same .json and append all messages again — duplicating
whatever was written before the crash.
Switch to SetHistory which atomically replaces the session contents.
A retry after crash overwrites the partial data instead of appending.
In SetHistory and Compact, the JSONL file was rewritten before updating
the meta file. If the process crashed between the two writes, the meta
still had a large Skip value pointing past the now-shorter JSONL file,
causing GetHistory to return empty — effectively data loss.
Reverse the order: write meta (with Skip=0) first, then rewrite JSONL.
On crash between the two writes, the old uncompacted file is still
intact and GetHistory reads from line 1, returning stale-but-complete
data. The next operation self-corrects.
A crash between the JSONL append and the meta update in addMsg can
leave meta.Count stale (e.g. file has 101 lines but meta says 100).
The previous code only reconciled when Count==0, so a nonzero stale
count was silently trusted, causing keepLast/skip to be calculated
against the wrong total.
Now TruncateHistory always counts the actual lines on disk. This is
cheap (scan without unmarshal) and TruncateHistory is not a hot path.
Implement TypingCapable interface for LINE channel using the
loading animation API (1:1 chats only, no group support).
- Add StartTyping() with 50s periodic refresh and context-based stop
- Integrate PlaceholderRecorder.RecordTypingStop in processEvent
- Skip RecordPlaceholder (LINE has no message edit API)
- Change sendLoading to accept context and return error
- Relax callAPI status check from 200 to 2xx range
Design consulted with Codex (GPT-5.2).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Address feedback from @yinwm for long-running daemon use:
- Replace sync.Map with a fixed-size sharded lock array (64 mutexes).
Keys are mapped via FNV hash, so memory is O(1) regardless of how
many sessions are created over the process lifetime.
- Increase scanner buffer cap from 1 MB to 10 MB. Tool results
(read_file on large files, web search responses) can easily exceed
1 MB. The scanner still starts at 64 KB and only grows as needed.
Address Codex (GPT-5.2) review feedback:
- Start: guard against Interval<=0 or MaxAge<=0 to prevent
time.NewTicker panic on misconfiguration
- ReleaseAll: split into two phases (collect under lock, delete
after unlock) matching CleanExpired pattern
- ReleaseAll: log file removal errors
- Add TestStartZeroIntervalNoPanic and TestStartZeroMaxAgeNoPanic
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- CleanExpired: split into two phases — collect expired entries under
lock, then delete files after releasing the lock to minimize contention
- CleanExpired: guard against zero MaxAge (no-op if unconfigured)
- CleanExpired: log file removal errors instead of silently ignoring
- Start: protect with startOnce to prevent multiple goroutines
- Stop: rename once -> stopOnce for clarity
- cmd_gateway: call mediaStore.Stop() on error path after Start()
- Add TestCleanExpiredZeroMaxAge and double-Start test
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Address review feedback from @Zhaoyikaiii:
- Replace map[string]*sync.Mutex + separate mu with sync.Map.LoadOrStore
for simpler, lock-free session lock management.
- Add skip parameter to readMessages so callers (GetHistory, Compact)
can skip truncated lines without paying the json.Unmarshal cost.
- Add countLines helper for TruncateHistory's count reconciliation,
avoiding full deserialization when only the line count is needed.
Address file growth concern from #711 review: logical truncation via
skip offset is fast but leaves dead lines on disk indefinitely.
Compact() rewrites the JSONL file keeping only active messages, using
the same temp+rename pattern for crash safety. No-op when skip == 0.
The caller (lifecycle manager or agent loop) decides when to trigger
compaction — e.g. when skipped lines exceed active lines.
Read existing sessions/*.json files, convert to JSONL format, and
rename originals to .json.migrated as backup. The migration is
idempotent — second runs skip already-migrated files.
Session keys are read from JSON content (not filenames) so that
sanitized names like telegram_123 correctly map back to telegram:123.
Cover all Store interface methods plus edge cases:
- Basic roundtrip, ordering, empty session, tool calls
- Logical truncation (keep last N, keep zero, keep more than exist)
- SetHistory replacing all + resetting skip offset
- Crash recovery with partial JSON lines
- Persistence across store instances
- Concurrent add+read (10 goroutines x 20 msgs)
- Simulated #704 race (summarizer vs main loop)
- Benchmarks for AddMessage and GetHistory (100/1000 msgs)
Add JSONLStore that persists sessions as .jsonl files (one message per
line) plus .meta.json for summary and truncation offset.
Key design decisions:
- Append-only writes — no full-file rewrites on AddMessage
- Logical truncation via skip offset instead of physical deletion
- Per-session mutex for safe concurrent access
- Crash recovery: malformed trailing lines are silently skipped
- Atomic metadata writes using temp+rename
Zero new dependencies — pure stdlib.
Refs #711
Introduce a backend-agnostic Store interface in pkg/memory/ that maps
one-to-one with the current SessionManager API. Each method is atomic
— no separate Save() call needed.
Refs #711
* fix: remove redundant tools definitions from system prompt
Tools are already provided to the LLM via JSON schema through
ToProviderDefs(), so the text-based tools section in the system
prompt is redundant.
This removes the buildToolsSection() logic and the tools field
from ContextBuilder, reducing system prompt length while maintaining
the "ALWAYS use tools" rule reminder.
Fixes#731
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: correct spelling 'initialized' (was 'initialised')
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
