feat(wren): add memory list, forget, dump & load commands#1531
Open
goldmedal wants to merge 4 commits intoCanner:mainfrom
Open
feat(wren): add memory list, forget, dump & load commands#1531goldmedal wants to merge 4 commits intoCanner:mainfrom
goldmedal wants to merge 4 commits intoCanner:mainfrom
Conversation
github-actions
bot
added
the
dependencies
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
🚧 Files skipped from review as they are similar to previous changes (1)
📝 WalkthroughWalkthroughAdds query-pair management: new MemoryStore query-history APIs, CLI subcommands to list/forget/dump/load pairs with YAML import/export and timestamps, scaffolded Changes
Sequence Diagram(s)sequenceDiagram
actor User
participant CLI as "CLI"
participant Inquirer as "InquirerPy"
participant Store as "MemoryStore"
participant DB as "LanceDB/Storage"
alt Non-interactive ID (--id)
User->>CLI: memory forget --id 3 [--force]
CLI->>Store: forget_queries_by_ids([3])
Store->>DB: DELETE rows by positional index
DB-->>Store: deleted count
Store-->>CLI: deleted count
else Source + Force (--source --force)
User->>CLI: memory forget --source appX --force
CLI->>Store: forget_queries_by_source("appX")
Store->>DB: DELETE where tags LIKE "source:appX"
DB-->>Store: deleted count
Store-->>CLI: deleted count
else Interactive
User->>CLI: memory forget
CLI->>Store: list_queries()
Store->>DB: SELECT paginated pairs
DB-->>Store: pairs
Store-->>CLI: pairs
CLI->>Inquirer: prompt_checkbox(pairs)
Inquirer-->>User: display & collect selection
User->>Inquirer: selects
Inquirer-->>CLI: selected IDs
CLI->>Store: forget_queries_by_ids(selected IDs)
Store->>DB: DELETE rows by positional index
DB-->>Store: deleted count
Store-->>CLI: deleted count
end
CLI-->>User: report result
sequenceDiagram
actor User
participant CLI as "CLI"
participant YAML as "yaml.safe_load / yaml.dump"
participant Store as "MemoryStore"
participant DB as "LanceDB/Storage"
rect rgba(100, 150, 200, 0.5)
Note over CLI,YAML: Load workflow
User->>CLI: memory load queries.yml [--upsert/--overwrite/--dry-run]
CLI->>YAML: parse file -> pairs
YAML-->>CLI: pairs list
CLI->>Store: load_queries(pairs, overwrite, upsert, dry_run)
Store->>DB: SELECT existing pairs
DB-->>Store: existing rows
Store->>Store: _existing_pairs_index() (dedupe/upsert mapping)
alt dry_run
Store-->>CLI: summary (loaded/updated/skipped)
else apply changes
Store->>DB: INSERT/DELETE as needed
DB-->>Store: result counts
Store-->>CLI: loaded/updated/skipped counts
end
end
rect rgba(150, 200, 100, 0.5)
Note over CLI,YAML: Dump workflow
User->>CLI: memory dump [--source X] [--output file]
CLI->>Store: dump_queries(source)
Store->>DB: SELECT pairs (omit vector)
DB-->>Store: pairs
Store-->>CLI: pairs
CLI->>YAML: _pairs_to_yaml(pairs)
YAML-->>CLI: YAML string
alt output specified
CLI->>FS: write to file or stdout
else default
CLI->>FS: write to project/queries.yml
end
CLI-->>User: done
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
goldmedal
force-pushed
the
feat/memory-list-forget-dump-load
branch
from
91b9707 to
99366b9
Compare
![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: 1
🧹 Nitpick comments (4)
wren/src/wren/memory/store.py (2)
372-392: Table rebuild on delete is acceptable but worth noting.The
forget_queries_by_idsimplementation drops and recreates the entire table to delete rows. This is simple and correct, but:
- Loses any LanceDB-specific metadata or indexes
- Not efficient for large tables
For the expected use case (small number of curated queries), this is likely fine. If this becomes a bottleneck, consider using LanceDB's native delete capability if available in newer versions.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@wren/src/wren/memory/store.py` around lines 372 - 392, forget_queries_by_ids currently drops and recreates the entire _QUERY_TABLE (using self._db.drop_table and create_table with _query_table_schema), which loses LanceDB-specific metadata/indexes and is inefficient for large tables; modify this function to use LanceDB's native row deletion API (or incremental delete methods) to remove rows by their positional IDs while preserving table metadata and indexes, and only fall back to the rebuild approach when the LanceDB version/engine doesn't support native deletes.
473-478: O(n²) complexity when upserting multiple rows.The
load_queriesupsert path calls_existing_pairs_index()after each deletion, rebuilding the entire index. Fornupserts, this results in O(n²) complexity.Consider collecting all row IDs to delete first, then performing a single batch delete, then inserting all new rows:
♻️ Suggested optimization for batch upsert
if upsert: - if nl in nl_to_rowid: - self.forget_queries_by_ids([nl_to_rowid[nl]]) - # Re-build index after deletion (row positions shift). - _, nl_to_rowid = self._existing_pairs_index() - updated += 1 - else: - loaded += 1 + # First pass: collect rows to delete + ids_to_delete = [] + for p in pairs: + nl = p["nl"] + if nl in nl_to_rowid: + ids_to_delete.append(nl_to_rowid[nl]) + + # Batch delete + if ids_to_delete: + self.forget_queries_by_ids(ids_to_delete) + updated = len(ids_to_delete) + + # Insert all pairs + for p in pairs: + tags = f"source:{p.get('source', 'user')}" + self.store_query( + nl_query=p["nl"], + sql_query=p["sql"], + datasource=p.get("datasource"), + tags=tags, + ) + loaded = len(pairs) - updated + return {"loaded": loaded, "skipped": 0, "updated": updated}🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@wren/src/wren/memory/store.py` around lines 473 - 478, The upsert path in load_queries is rebuilding the full _existing_pairs_index() after every single forget_queries_by_ids call, causing O(n²) behavior; instead, collect all row IDs to remove by looking up nl in nl_to_rowid, call forget_queries_by_ids once with the batch of IDs, then call _existing_pairs_index() a single time before performing the new inserts and updating updated; update the logic around nl_to_rowid, forget_queries_by_ids, and the section that increments updated to operate on the batched delete result rather than per-item.wren/src/wren/memory/cli.py (2)
489-494: Edge case:_parse_sourcemay not handle multi-word tags correctly.If
tagscontains multiple space-separated tags like"source:user verified", the function works correctly. However, if someone stores a tag like"verified source:user"without spaces (e.g.,"verified,source:user"), it would fail to extract the source.Given that the store itself uses the format
f"source:{...}"consistently, this is unlikely to be an issue in practice, but worth noting.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@wren/src/wren/memory/cli.py` around lines 489 - 494, _parse_source currently only splits tags by whitespace and misses cases like "verified,source:user"; update the function to robustly find a source token anywhere in the tags string (e.g., use a regex search for a word boundary followed by "source:" and capture the value, such as r'\bsource:([^,\s]+)') and return that capture if found, otherwise return "user". Modify the _parse_source implementation to use this search approach so it works with comma-separated or mixed delimiters while keeping the same return behavior.
161-220: Consider narrowing the exception handling for queries.yml auto-load.The broad
except (SystemExit, Exception)at line 219 silently swallows all errors, including potential bugs likeKeyErrororTypeErrorin the load logic. While best-effort loading is the intent, completely hiding errors makes debugging difficult.Consider logging a debug message or at least catching more specific exceptions:
♻️ Suggested improvement
try: from wren.context import discover_project_path # noqa: PLC0415 project_path = discover_project_path(explicit=None) queries_file = project_path / "queries.yml" if queries_file.exists(): raw = queries_file.read_text(encoding="utf-8") doc = yaml.safe_load(raw) if doc and isinstance(doc, dict) and doc.get("pairs"): load_result = mem_store.load_queries(doc["pairs"], upsert=False) loaded = load_result["loaded"] skipped = load_result["skipped"] if loaded: typer.echo( f"Loaded {loaded} pair(s) from queries.yml" f" ({skipped} skipped).", err=True, ) - except (SystemExit, Exception): - pass # queries.yml loading is best-effort + except (SystemExit, FileNotFoundError, PermissionError, yaml.YAMLError): + pass # queries.yml loading is best-effort; ignore known failure modes🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@wren/src/wren/memory/cli.py` around lines 161 - 220, The final broad except (SystemExit, Exception) around the queries.yml auto-load should be replaced with narrower, explicit exception handling: catch import/IO errors (e.g., FileNotFoundError, PermissionError, IsADirectoryError, UnicodeDecodeError), YAML parsing errors (yaml.YAMLError), and the specific runtime errors you expect from parsing/loading (e.g., KeyError, TypeError, ValueError) instead of Exception; when one of these occurs, log a brief debug/error message including the exception details (use typer.echo(..., err=True) or the project's logger) so failures are visible but do not fail the command; update the try/except that wraps discover_project_path, queries_file.read_text, yaml.safe_load, and mem_store.load_queries and include the exception variable in the log to aid debugging while preserving best-effort behavior.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@wren/src/wren/memory/store.py`:
- Around line 353-362: list_queries currently resets the DataFrame index
(reset_index(drop=True)) so the assigned _row_id (offset + i) reflects the
position in the filtered/sorted view, which no longer matches the original table
indices used by forget_queries_by_ids; change list_queries to preserve the
original DataFrame index when sorting (e.g., use df.sort_values("created_at",
ascending=False).reset_index() or otherwise retain the original index column),
then when building results set each record's _row_id from that original index
value (the preserved index column) instead of offset + i so the IDs align with
the unfiltered table used by forget_queries_by_ids.
---
Nitpick comments:
In `@wren/src/wren/memory/cli.py`:
- Around line 489-494: _parse_source currently only splits tags by whitespace
and misses cases like "verified,source:user"; update the function to robustly
find a source token anywhere in the tags string (e.g., use a regex search for a
word boundary followed by "source:" and capture the value, such as
r'\bsource:([^,\s]+)') and return that capture if found, otherwise return
"user". Modify the _parse_source implementation to use this search approach so
it works with comma-separated or mixed delimiters while keeping the same return
behavior.
- Around line 161-220: The final broad except (SystemExit, Exception) around the
queries.yml auto-load should be replaced with narrower, explicit exception
handling: catch import/IO errors (e.g., FileNotFoundError, PermissionError,
IsADirectoryError, UnicodeDecodeError), YAML parsing errors (yaml.YAMLError),
and the specific runtime errors you expect from parsing/loading (e.g., KeyError,
TypeError, ValueError) instead of Exception; when one of these occurs, log a
brief debug/error message including the exception details (use typer.echo(...,
err=True) or the project's logger) so failures are visible but do not fail the
command; update the try/except that wraps discover_project_path,
queries_file.read_text, yaml.safe_load, and mem_store.load_queries and include
the exception variable in the log to aid debugging while preserving best-effort
behavior.
In `@wren/src/wren/memory/store.py`:
- Around line 372-392: forget_queries_by_ids currently drops and recreates the
entire _QUERY_TABLE (using self._db.drop_table and create_table with
_query_table_schema), which loses LanceDB-specific metadata/indexes and is
inefficient for large tables; modify this function to use LanceDB's native row
deletion API (or incremental delete methods) to remove rows by their positional
IDs while preserving table metadata and indexes, and only fall back to the
rebuild approach when the LanceDB version/engine doesn't support native deletes.
- Around line 473-478: The upsert path in load_queries is rebuilding the full
_existing_pairs_index() after every single forget_queries_by_ids call, causing
O(n²) behavior; instead, collect all row IDs to remove by looking up nl in
nl_to_rowid, call forget_queries_by_ids once with the batch of IDs, then call
_existing_pairs_index() a single time before performing the new inserts and
updating updated; update the logic around nl_to_rowid, forget_queries_by_ids,
and the section that increments updated to operate on the batched delete result
rather than per-item.
🪄 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: defaults
Review profile: CHILL
Plan: Pro
Run ID: b4f0daaf-45e7-4d9c-8ce4-bc3e4ee2d762
⛔ Files ignored due to path filters (1)
wren/uv.lockis excluded by!**/*.lock
📒 Files selected for processing (5)
wren/pyproject.tomlwren/src/wren/context_cli.pywren/src/wren/memory/cli.pywren/src/wren/memory/store.pywren/tests/unit/test_memory.py
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@wren/src/wren/memory/store.py`:
- Around line 466-485: The loop builds exact_set once from existing rows but
never updates it when new rows are inserted, so duplicate pairs in the incoming
pairs list are all inserted; update exact_set (and nl_to_rowid when appropriate)
immediately after inserting a new pair so subsequent iterations see the newly
inserted pair: inside the non-upsert branch where you do loaded += 1, add
exact_set.add((nl, sql)) after the insert; similarly in the upsert branch when
you insert (the path where nl was not in nl_to_rowid and you increment loaded)
add exact_set.add((nl, sql)) and nl_to_rowid[nl] = new_row_id (or call
_existing_pairs_index() if you prefer) so duplicates in the same input batch are
detected and skipped.
- Line 399: The WHERE clause in wren/src/wren/memory/store.py currently uses a
loose LIKE pattern that can match substrings; update the assignment of the local
variable where (the line using f"tags LIKE '%source:{_esc(source)}%'") to use an
exact tag match consistent with list_queries — i.e., match the full tag token
delimited by tag boundaries (commas/starts/ends) rather than any substring.
Locate the assignment to where in the function that builds the query and replace
the pattern so it only selects rows where a tag equals "source:{source}" (or
matches as a complete token using leading/trailing delimiters), preserving the
use of _esc(source) for escaping.
- Around line 350-351: The current filter df =
df[df["tags"].str.contains(f"source:{source}", na=False)] does substring
matching and will match unintended sources; change the filter on the "tags"
column to match the exact source token (e.g., use a regex with boundaries and
escape the source value or split the comma-separated tags and check membership)
so only tags like "source:<exact>" match; update the expression that references
df["tags"].str.contains(...) accordingly (preserve df and "tags" column usage).
🪄 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: defaults
Review profile: CHILL
Plan: Pro
Run ID: 0e959a77-b6ac-41dc-88ab-688b4797598c
⛔ Files ignored due to path filters (1)
wren/uv.lockis excluded by!**/*.lock
📒 Files selected for processing (5)
wren/pyproject.tomlwren/src/wren/context_cli.pywren/src/wren/memory/cli.pywren/src/wren/memory/store.pywren/tests/unit/test_memory.py
✅ Files skipped from review due to trivial changes (4)
- wren/src/wren/context_cli.py
- wren/pyproject.toml
- wren/tests/unit/test_memory.py
- wren/src/wren/memory/cli.py
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (1)
wren/src/wren/memory/cli.py (1)
393-403: Consider using_parse_sourcefor consistency.The source extraction logic here uses substring matching (
"source:seed" in tags) while_parse_source(line 501-506) parses more robustly by splitting on whitespace. For consistency and to handle edge cases, consider reusing_parse_source:def _format_choice_label(row: dict, max_nl: int = 40, max_sql: int = 50) -> str: """Format a row as a human-readable choice label.""" - source = "user" - tags = row.get("tags", "") - if "source:seed" in tags: - source = "seed" - elif "source:view" in tags: - source = "view" + source = _parse_source(row.get("tags", "")) nl = row.get("nl_query", "")[:max_nl]🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@wren/src/wren/memory/cli.py` around lines 393 - 403, _format_choice_label currently determines source by substring checks on the tags string; replace that ad-hoc logic with a call to the existing _parse_source helper to ensure consistent parsing: call _parse_source(row.get("tags", "")) (or pass tags variable) and use its return value (falling back to "user" if it returns falsy) instead of the manual "source:seed" / "source:view" checks; keep the rest of the function unchanged so nl and sql slicing behavior remains intact.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@wren/src/wren/memory/cli.py`:
- Around line 393-403: _format_choice_label currently determines source by
substring checks on the tags string; replace that ad-hoc logic with a call to
the existing _parse_source helper to ensure consistent parsing: call
_parse_source(row.get("tags", "")) (or pass tags variable) and use its return
value (falling back to "user" if it returns falsy) instead of the manual
"source:seed" / "source:view" checks; keep the rest of the function unchanged so
nl and sql slicing behavior remains intact.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 0d084560-fc36-4071-990b-c47db8e9875c
📒 Files selected for processing (2)
wren/src/wren/memory/cli.pywren/src/wren/memory/store.py
🚧 Files skipped from review as they are similar to previous changes (1)
- wren/src/wren/memory/store.py
Enable browsing, deleting, exporting, and importing NL-SQL pairs in the LanceDB query_history. This gives users precise control over memory quality instead of relying solely on `memory reset`. New CLI commands: - `wren memory list` — browse pairs with --source, --limit, --offset - `wren memory forget` — interactive (InquirerPy) or non-interactive (--id/--source --force) deletion - `wren memory dump` — export pairs to YAML (project queries.yml or stdout) - `wren memory load FILE` — import from YAML with skip/upsert/overwrite modes Project integration: - `context init` scaffolds queries.yml - `memory index` auto-loads queries.yml (--no-queries to skip) - `memory dump` defaults to project queries.yml when inside a project Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fix _row_id mismatch: preserve original DataFrame index in list_queries so forget_queries_by_ids deletes the correct rows when source filter is applied - Use exact tag match (== / =) instead of substring matching (contains / LIKE) across all source-filtering methods - Prevent duplicate pairs within input list during default load mode by updating exact_set after each insert - Batch upsert: collect all IDs to delete first, single batch delete, then insert all — eliminates O(n²) index rebuilding - Narrow exception handling in queries.yml auto-load to explicit types Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…es.yml Add documentation for the new memory management commands and project integration workflow: browsing pairs, interactive/non-interactive forget, YAML dump/load, queries.yml scaffolding, auto-load on index, agent hygiene tips, and full command reference table. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
goldmedal
force-pushed
the
feat/memory-list-forget-dump-load
branch
from
db5afe2 to
f70985d
Compare
github-actions
bot
added
the
documentation
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@docs/guide/memory.md`:
- Around line 234-244: Add a language specifier to the fenced code block that
shows the directory tree (the block starting with triple backticks and the
project_root/ tree) by changing the opening fence from ``` to ```text (or
```plaintext) so the markdown linter recognizes it as a non-code block; leave
the block contents unchanged and only update the opening fence.
In `@wren/src/wren/memory/store.py`:
- Around line 470-489: The upsert block inserts duplicates when the input pairs
contains repeated nl_query values because nl_to_rowid only covers existing rows;
fix by deduplicating or tracking during the upsert: before deleting/ inserting,
build a map from nl (use p["nl"] / nl_query) to the last pair (or iterate pairs
and keep the last occurrence) so you only delete once and insert once per nl, or
alternatively maintain an inserted_nl set inside the insertion loop (around
store_query calls) to skip subsequent pairs with the same nl; update references
to nl_to_rowid, ids_to_delete, and the store_query loop accordingly (use the
deduped_pairs map or inserted_nl set to compute updated/loaded counts
correctly).
🪄 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: defaults
Review profile: CHILL
Plan: Pro
Run ID: db25b879-dddc-46d4-bbb2-2d1cab3677e1
⛔ Files ignored due to path filters (1)
wren/uv.lockis excluded by!**/*.lock
📒 Files selected for processing (6)
docs/guide/memory.mdwren/pyproject.tomlwren/src/wren/context_cli.pywren/src/wren/memory/cli.pywren/src/wren/memory/store.pywren/tests/unit/test_memory.py
✅ Files skipped from review due to trivial changes (1)
- wren/src/wren/context_cli.py
🚧 Files skipped from review as they are similar to previous changes (2)
- wren/tests/unit/test_memory.py
- wren/src/wren/memory/cli.py
- Upsert mode now deduplicates input pairs by nl_query (last wins) before batch delete + insert, preventing duplicate rows when the same nl_query appears multiple times in the input - Add language specifier to fenced code block in memory docs Co-Authored-By: Claude Opus 4.6 (1M context) <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
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.
Summary
wren memory listto browse stored NL-SQL pairs with--source,--limit,--offset,--outputflagswren memory forgetwith interactive checkbox UI (InquirerPy) and non-interactive modes (--id,--source --force) for precise pair deletionwren memory dumpto export pairs as YAML (defaults to projectqueries.yml, or stdout with-o -)wren memory load FILEto import from YAML with three modes: skip-duplicates (default),--upsert,--overwritequeries.ymlinto project lifecycle: scaffolded bycontext init, auto-loaded bymemory index(skippable via--no-queries)interactiveoptional dependency group (InquirerPy>=0.3.4) with graceful fallbackTest plan
just test-memory)just lint)wren memory list,wren memory list --source seed --output jsonwren memory forget --id 0 --force,wren memory forget --source seed --forcewren memory dump -o queries.ymlthenwren memory load queries.ymlwren memory load queries.yml --dry-run,--upsert,--overwritewren context initcreatesqueries.ymlwren memory indexauto-loadsqueries.yml🤖 Generated with Claude Code
Summary by CodeRabbit