Bug: summarizer.py context_type misclassifies memories as resources during reindex

[muxiaomu001]

Bug Description

During reindex (via /api/v1/content/reindex), all memory files under viking://user/memories/ and viking://agent/memories/ are incorrectly tagged with context_type = "resource" instead of "memory".

This causes downstream consumers (e.g. OpenClaw's auto-recall plugin) that filter by context_type == "memory" to miss all reindexed memories.

Root Cause

In openviking/utils/summarizer.py (line ~61), the URI check uses a prefix that doesn't match actual memory paths:

# Current (broken)
if uri.startswith("viking://memory/"):
    context_type = "memory"
elif uri.startswith("viking://agent/skills/"):
    context_type = "skill"

Actual memory URIs look like:

• viking://user/memories/entities/mem_xxx.md
• viking://agent/<id>/memories/cases/mem_xxx.md

None of these start with viking://memory/, so they all fall through to the default "resource".

Fix

Use substring matching consistent with core/directories.py:get_context_type_for_uri():

# Fixed
if "/memories" in uri:
    context_type = "memory"
elif "/skills" in uri:
    context_type = "skill"

Note: core/directories.py already has the correct logic ("/memories" in uri). The summarizer should match.

Impact

• ov find works fine (no type filter)
• OpenClaw auto-recall (context_type == "memory") misses all reindexed memories
• New memories captured via auto-capture are unaffected (different code path)

Versions Affected

Confirmed in 0.2.9 and 0.2.13.

Workaround

sed -i 's|if uri.startswith("viking://memory/")|if "/memories" in uri|' \
  $(python3 -c "import openviking; print(openviking.__path__[0])")/utils/summarizer.py
sed -i 's|elif uri.startswith("viking://agent/skills/")|elif "/skills" in uri|' \
  $(python3 -c "import openviking; print(openviking.__path__[0])")/utils/summarizer.py

Then re-run reindex:

curl -X POST http://127.0.0.1:1933/api/v1/content/reindex \
  -H "Content-Type: application/json" \
  -d '{"uri": "viking://", "regenerate": true, "wait": false}'

[muxiaomu001]

Update: Second related bug in semantic_dag.py

After further investigation, there is a second bug in the same code path that compounds the first one.

Bug 2: SemanticDagExecutor propagates root-level context_type to all child nodes

File: openviking/storage/queuefs/semantic_dag.py
Lines: 443 and 563

When SemanticDagExecutor is initialized, it receives context_type from the SemanticMsg (which represents the root URI being reindexed). It stores this as self._context_type and uses it for every child node during recursive traversal:

# Line 443 (file vectorization)
task = VectorizeTask(
    task_type="file",
    uri=file_path,
    context_type=self._context_type,  # ← always uses root's type
    ...
)

# Line 563 (directory vectorization)
task = VectorizeTask(
    task_type="directory",
    uri=dir_uri,
    context_type=self._context_type,  # ← always uses root's type
    ...
)

When reindexing from viking:// (root), the root URI has no /memories substring, so context_type defaults to "resource". This "resource" type then propagates to all child nodes — including memories under viking://user/*/memories/*.

Fix

Use the existing get_context_type_for_uri() from core/directories.py to determine each node's type individually:

from openviking.core.directories import get_context_type_for_uri

# Line 443
context_type=get_context_type_for_uri(file_path),

# Line 563
context_type=get_context_type_for_uri(dir_uri),

This is consistent with how DirectoryInitializer determines context types.

Why this bug is hidden

Normal memory creation via auto-capture goes through memory_updater.py → embedding queue (which hardcodes context_type="memory"). It never touches Summarizer or SemanticDagExecutor. So this bug only manifests during reindex operations — typically triggered after switching embedding models.

Summary of both bugs

#	File	Line	Issue	Fix
1	summarizer.py	61	uri.startswith("viking://memory/") never matches real URIs	Change to "/memories" in uri
2	semantic_dag.py	443, 563	Root URI's context_type propagated to all children	Use get_context_type_for_uri(uri) per node

Both bugs must be fixed together for reindex to correctly classify memories.

[deepakdevp]

I'd like to take this. The fix is straightforward — replace the broken uri.startswith("viking://memory/") check in summarizer.py with a call to the existing get_context_type_for_uri() from core/directories.py, which already uses the correct "/memories" in uri substring matching. Will include tests. PR incoming shortly.