lope-memory

name

lope-memory

description

Query the persistent Lope finding store (SQLite, populated by `lope review --remember`). Use when the user asks 'is this a recurring issue?', 'which files keep getting flagged?', 'show me everything Lope knows about auth.py', 'what was that finding from last week?', or 'how many findings has Lope stored?'. Five subcommands: stats, search, file, hotspots, forget. All local, stdlib only, opt-in — `LOPE_MEMORY=off` disables the store entirely.

Lope Memory

lope review --remember upserts every consensus finding into a local SQLite store at ~/.lope/memory.db (override via LOPE_MEMORY_DB). The lope memory verb lets the user query that store.

Invocation

Two paths — you must handle both:

1. Explicit slash command. User types lope memory hotspots --days 30. Route directly.
2. Natural language. User says one of:
   • "What does Lope remember about auth.py?" → lope memory file auth.py
   • "Have we seen this before?" / "Is this a recurring issue?" → lope memory search "<keywords from the current finding>"
   • "Which files keep getting flagged?" → lope memory hotspots --days 30
   • "How many findings has Lope stored?" / "memory stats" → lope memory stats
   • "Forget the finding for src/legacy/old.py" → lope memory forget --file src/legacy/old.py

Command shape

lope memory stats                              # totals + recurring + last session
lope memory stats --json                       # machine-readable

lope memory search "rate limit"                # LIKE on message
lope memory search "csrf" --min-score 0.5      # filter by consensus_score
lope memory search "x" --limit 50

lope memory file path/to/file.py               # all findings for one file
lope memory file path/to/file.py --json --limit 100

lope memory hotspots                           # last 30 days, top 10
lope memory hotspots --days 7 --limit 20

lope memory forget --hash <16-char-hash>       # remove one finding
lope memory forget --file legacy/old.py        # nuke every finding for one file

How to populate the store

Memory is opt-in. The store stays empty until the user runs:

lope review <target> --consensus --remember

If lope memory stats shows total_findings: 0, suggest running a consensus review with --remember first.

Disabled-state behavior

LOPE_MEMORY=off (or no/0/false/disabled) disables the store. In that state:

• lope review --remember prints a visible no-op note: Memory disabled (LOPE_MEMORY=off); --remember was a no-op.
• lope memory <subcmd> exits 2 with Lope memory is disabled (LOPE_MEMORY=off). Unset the variable or set LOPE_MEMORY= to re-enable.

If the user encounters either message and wants memory back, they should unset LOPE_MEMORY (bash) or open a fresh shell.

Redaction

Every text path through memory passes through lope.redaction.redact_text before any column is bound. Bearer tokens, sk-* keys, GitHub PATs, and PEM blocks never reach disk. This is a docstring guarantee in lope/memory.py and is regression-tested.

Hard rules

• No external service. The store is a local SQLite file. Nothing runs in the background.
• No mutations on read. stats, search, file, hotspots are read-only.
• forget requires a selector. Running lope memory forget with no --hash or --file is a usage error, not a wipe.
• Recurring-issue UX. When a --remember run encounters a finding hash that's already stored, seen_count increments and first_seen_at is preserved. The CLI footer surfaces "X recurring" so users see when something keeps coming back.

Composability

Pair memory with other v0.7 verbs:

# Surface recurring issues during PR review
lope review pr.diff --divide hunks --consensus --remember
lope memory hotspots --days 14

# Brain-aware persistent reviews (Makakoo OS)
lope review auth.py --consensus --remember --brain-context "auth decisions" --brain-log

For full details: lope docs (docs/memory.md).