feat(session): add batch memory compression CLI command

[mvanhorn]

Problem Statement

Memory directories grow without bound as agents run over time. Verbose abstracts waste tokens during overview generation. A user in #350 reported 150K tokens for a single overview because abstracts are longer than the source memories themselves.

Related: #350 (decoupling ingestion), #578 (custom prompts), RFC #712 (memory templating).

Proposed Solution

Add ov compress CLI command that scans a directory for memories with verbose abstracts and truncates them to a target length.

ov compress viking://user/memories/ --max-abstract-length 128 --dry-run

Evidence

Source	Evidence	Engagement
#350	Decoupling ingestion from summarization	3 thumbsup
#350 comment	150K tokens per overview from verbose abstracts	direct user report
#578	Prompt template customization demand	2 thumbsup, @qin-ctx engaged
Prism MCP	10x memory compression demand	64 upvotes, 19 comments

Changes

• openviking/utils/compress_service.py (new): CompressService scans directory, filters memories by abstract length, truncates excess
• openviking/server/routers/content.py: POST /v1/content/compress endpoint following the reindex pattern
• crates/ov_cli/src/client.rs: compress() HTTP client method
• crates/ov_cli/src/commands/content.rs: compress command handler
• crates/ov_cli/src/main.rs: Compress subcommand with --max-abstract-length and --dry-run flags

Usage

# Preview what would be compressed (no changes made)
ov compress viking://user/memories/ --dry-run

# Compress abstracts exceeding 128 chars (default)
ov compress viking://user/memories/

# Custom threshold
ov compress viking://user/memories/ --max-abstract-length 256

Returns: files_scanned, files_compressed, chars_saved, and a list of verbose files.

Testing

3 Python unit tests covering CompressService init and error handling.

tests/unit/utils/test_compress_service.py ...  [100%]
3 passed

Implementation Notes

• Follows the same CLI -> HTTP -> service pattern as ov reindex (merged in PR feat(cli): add reindex command to trigger content re-indexing #795)
• --dry-run returns stats without modifying anything, safe to run anytime
• Truncation uses word-boundary splitting to avoid cutting mid-word
• Caps verbose file list at 20 entries in the response to avoid payload bloat

Feature Area

Session Management

This contribution was developed with AI assistance (Claude Code).

[CLAassistant]

All committers have signed the CLA.

[github-actions]

Failed to generate code suggestions for PR