Help us improve
Share bugs, ideas, or general feedback.
From rldyour-serena-mcp
Синхронизация .serena/memories - fact-only заметки из кода/диффов/тестов. Используй для: обнови memories, синхронизируй память, после коммита, после Stop-хука, .serena/plans, .serena/research, долгоживущие изменения. EN triggers: sync memories, refresh project knowledge, fact-only memory audit, post-commit sync, post-Stop sync, durable changes, update Serena memories.
npx claudepluginhub nddev-it-com/rldyour-claudecode --plugin rldyour-serena-mcpHow this skill is triggered — by the user, by Claude, or both
Slash command
/rldyour-serena-mcp:serena-memory-syncThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Keep `.serena/` useful for future Claude Code, Codex, and other GPT-based agent sessions without creating hallucinated project lore. Memories must explain what the code actually does, where it lives, how it behaves, which invariants matter, and how to safely change and verify the area.
Guides technical evaluation of code review feedback: read fully, restate for understanding, verify against codebase, respond with reasoning or pushback before implementing.
Share bugs, ideas, or general feedback.
Keep .serena/ useful for future Claude Code, Codex, and other GPT-based agent sessions without creating hallucinated project lore. Memories must explain what the code actually does, where it lives, how it behaves, which invariants matter, and how to safely change and verify the area.
User-facing conversation stays in Russian. All stored .serena/ knowledge files are written in English.
Use this skill without waiting for explicit invocation when:
.serena/memories, .serena/plans, or .serena/research may be stale or need a fact-only audit.Do not auto-invoke this skill for read-only context discovery, log/server audits, report-only reviews, exploratory debugging, or current-status snapshots unless the user explicitly asked to update/synchronize memories or a Stop/stale-memory hook requires it. In those read-only workflows, list useful candidates in the user report and wait for permission before writing .serena.
Do not create memory noise for trivial formatting, purely mechanical edits, current runtime status, timestamped snapshots, server log summaries, health-check output, or unverified assumptions.
.serena/memories/: durable project facts, conventions, architecture notes, implementation facts, and task completion requirements..serena/plans/: non-trivial implementation plans that are worth preserving across sessions..serena/research/: complex or long research results with source links and implementation impact.In normal product repositories, these knowledge files are agent-only context. They are restored from and published to fullrepo, then ignored through .git/info/exclude instead of being committed to main. Repositories that are themselves agent tooling may intentionally track selected .serena knowledge files when they are part of the product source of truth.
Local/runtime files must not be committed or published by this plugin: .serena/cache/, .serena/.gitignore, .serena/project.local.yml, .serena/.sync_marker, .serena/.serena_sync_state.json, .serena/.auto_sync_head, .serena/.active_workflow_intent.json, .serena/.dirty_stop_ack, .serena/.flow_sync_marker, .serena/.flow_post_task_state.json.
Name memory files as AREA-01-SLUG.md.
Default cross-project areas: CORE, BACKEND, FRONTEND, MOBILE, INFRA, API, AUTH, DATA, SEC, TEST, DESIGN, CLI, MCP, DOCS, RELEASE, TECHDEBT. Agent-tooling repositories may add tool-specific areas such as CLAUDECODE when they are clearer than overloading CORE.
Create custom area prefixes only when the project needs a clearer domain boundary. Keep files narrow and split large memories instead of creating broad catch-all files.
Every project should maintain a compact index memory, normally CORE-01-INDEX.md, that lists the active memory areas, canonical file names, and which source-of-truth paths each memory owns. When adding a new memory, update the index in the same sync pass.
Numbering is stable:
01 is normally the area index or the first canonical topic.02, 03, ... for new durable topics inside the same area.SERENA-01-MEMORY-SYNC.md, TECHDEBT-01-NOW.md.Split a memory when it starts covering more than one domain responsibility, when sections become hard to scan, or when a future agent would need to read many unrelated facts to answer one scoped question. Prefer adding a specific numbered file over appending unrelated content to an older broad memory.
Do not create timestamped snapshot memories such as SYNC_YYYYMMDD_*. If a runtime observation reveals a durable contract, update the relevant area memory with the stable contract and source paths instead of storing the observation as a snapshot.
Every memory starts with exactly this metadata block:
<!-- Memory Metadata
Last updated: YYYY-MM-DD
Last commit: <sha> <message>
Scope: <files/dirs>
Area: <AREA>
-->
Do not add subjective confidence fields or unsupported metadata. If a fact is unresolved, put it in the body as an explicit unresolved gap.
After the metadata block, use this body structure unless a section is truly irrelevant. Keep sections concise and factual.
# <AREA-01-SLUG>
## Purpose
What this area is responsible for in the current project.
## Source Of Truth
- `path/to/file`: verified role of the file.
- `path/to/directory/`: verified role of the directory.
## Entry Points
- `symbol_or_command`: when it is used and what it controls.
## Current Behavior
Facts about runtime behavior, data flow, configuration flow, or lifecycle behavior that were verified from code, git diff, tests, or authoritative project files.
## Contracts And Data
File formats, config keys, environment variables, schemas, API shapes, command arguments, naming conventions, and persistence rules that future changes must preserve.
## Invariants
- Conditions that must remain true after changes.
- Boundaries that must not be crossed.
## Change Rules
- How to modify this area safely.
- Which tools or project patterns should be used first.
## Verification
- `command`: what it proves.
- Manual check: exact behavior to inspect if no command exists.
## Known Gaps
- Unverified or intentionally unresolved facts. Omit this section when there are no gaps.
Use exact file paths and symbol names. If a section would only contain generic advice, omit it rather than adding filler.
Capture facts that materially improve future implementation confidence:
Do not write:
.serena/plans for current non-trivial plans.python3 plugins/rldyour-serena-mcp/scripts/serena_memory_state.py (state + stale marker)jq/python3 read .serena/.serena_sync_state.json if present; use analysis.memory_taxonomy, analysis.memory_targets, analysis.areas, and analysis.areas_summary as the first-pass scope.analysis.schema_version as the analyzer payload contract; if the field is missing, proceed conservatively from changed files.sync_state.changed_files when available.analysis.areas and analysis.memory_targets.changed_files_since_sync in state.list_memories, then read CORE-01-INDEX when present and every relevant memory target inferred from names. For code, use get_symbols_overview, targeted find_symbol, and find_referencing_symbols before raw reads when the language server supports the file type..serena/plans/ only when they will help future sessions continue work..serena/research/ only when the research was complex, source-backed, and likely reusable.commit_serena_knowledge.sh script (path provided by the Stop hook message, or ${CLAUDE_PLUGIN_ROOT}/scripts/commit_serena_knowledge.sh when the plugin is enabled). In repositories where .serena knowledge is still tracked, this creates the knowledge-only commit. In fullrepo-managed repositories, it acknowledges current memories and clears runtime sync markers without committing AI files to the current branch; flow-post-task-sync publishes the final fullrepo snapshot.Code is the source of truth. Never write a memory fact just because an old memory, plan, comment, or README says it.
Prefer small factual paragraphs over narratives. A future GPT-based coding agent should be able to quickly answer: what exists, where it is implemented, what conventions apply, what behavior and contracts matter, what must not break, how to safely change it, and what checks prove correctness.
Do not store secrets, tokens, private cookies, raw credentials, or sensitive runtime data.
Do not use .serena/memories as a backlog. Plans and TODOs belong in .serena/plans only when the plan is useful and current.
Report:
Freshness audit: HEAD, newest synced commit, and changed scope.Updated memories: edited files.New memories: new files, if any.Plans/research archived: files written, if any.Unresolved gaps: anything that could not be verified from code.Commit: whether the Serena knowledge-only auto-commit was created or fullrepo-managed knowledge was acknowledged for later fullrepo publish.