From hoangsa
Use when the user needs to run hoangsa-memory CLI commands — setup / index / query / watch / impact / context / changes / graph / memory / skills / eval / uninstall. Examples: "index this repo", "show memory", "trace the graph", "find taint paths", "uninstall hoangsa-memory".
How this skill is triggered — by the user, by Claude, or both
Slash command
/hoangsa:memory-cliThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Every hoangsa-memory MCP tool has a CLI equivalent, plus a few CLI-only
Every hoangsa-memory MCP tool has a CLI equivalent, plus a few CLI-only
commands (setup, watch, eval). Run from the repo root unless noted —
the CLI defaults --root to ./.hoangsa/memory.
hoangsa-memory setupOne-shot install. Writes ./.hoangsa/memory/config.toml, seeds MEMORY.md +
LESSONS.md, merges hoangsa-memory hooks + skills + MCP server into
.claude/settings.json (or ~/.claude/settings.json with
--scope user). Re-run any time to reconfigure / self-heal.
hoangsa-memory setup # interactive
hoangsa-memory setup --yes # accept defaults (CI / scripts)
hoangsa-memory setup --status # show install state, don't modify
hoangsa-memory uninstallRemoves hoangsa-memory's managed hooks + skills + MCP entry from
.claude/settings.json and .mcp.json. Leaves the .hoangsa/memory/ data
directory intact — delete it manually if you want a hard reset.
hoangsa-memory uninstall # project scope
hoangsa-memory uninstall --scope user # user scope
hoangsa-memory index [path]Parse + index a source tree. Populates chunks.db, the graph, and
(if --embedder is set) vectors.db.
hoangsa-memory index . # Mode::Zero — BM25 + symbol + graph
hoangsa-memory index . --embedder voyage # Mode::Full — adds semantic vectors
Embedders require their API key in the matching env var
(VOYAGE_API_KEY, OPENAI_API_KEY, COHERE_API_KEY) and the matching
Cargo feature at build time.
hoangsa-memory watch [path]Re-index on file save. Cheaper than re-running hoangsa-memory index manually
during an active session.
hoangsa-memory watch .
hoangsa-memory watch . --debounce-ms 500
hoangsa-memory query <text...>Hybrid recall. Joins extra args with spaces — no quoting needed for multi-word queries.
hoangsa-memory query authentication login session
hoangsa-memory query -k 16 retry pool exhausted # more hits
hoangsa-memory query --json error handler 500 # machine-readable
hoangsa-memory impact <fqn>Blast-radius analysis. Direction defaults to up (who calls this);
down is "what does this depend on"; both is the union.
hoangsa-memory impact server::dispatch_tool
hoangsa-memory impact auth::verify_token -d 5
hoangsa-memory impact util::fmt --direction down
hoangsa-memory context <fqn>360° view of a symbol: callers, callees, extends, extended_by, references, siblings, unresolved imports.
hoangsa-memory context server::dispatch_tool
hoangsa-memory context auth::Session --limit 64
hoangsa-memory changesChange-impact over a unified diff. With no --from, runs
git diff HEAD in the current tree.
hoangsa-memory changes # current working-tree diff
hoangsa-memory changes --from patch.diff # from a file
gh pr diff 123 | hoangsa-memory changes --from -
hoangsa-memory changes -d 3 # deeper upstream walk
Whole-shape queries over the code graph — reach for these instead of
grepping call chains by hand. All accept --json.
hoangsa-memory graph query <start...>Traverse callers/callees/refs/imports from seed symbol(s). Filter by direction / edge kinds / depth; export JSON or DOT.
hoangsa-memory graph query server::dispatch_tool --direction both -d 2
hoangsa-memory graph query auth::verify_token --edge-kinds calls --format dot
hoangsa-memory graph paths <from> <to>Shortest dependency/call path between two symbols.
hoangsa-memory graph paths router::handle_login db::users::find_by_email
hoangsa-memory graph communitiesArchitecture map — clusters of tightly-coupled symbols (largest first).
hoangsa-memory graph communities --min-size 5
hoangsa-memory graph processesExecution flows from entry points (::main or --entry-globs).
hoangsa-memory graph processes -d 8
hoangsa-memory graph taintSource→sink security dataflow. Requires an index built with
hoangsa-memory index . --pdg. Omit --source/--sink for built-in defaults.
hoangsa-memory index . --pdg
hoangsa-memory graph taint --source env::var --sink Command::new --json
hoangsa-memory memory showPrint MEMORY.md + LESSONS.md.
hoangsa-memory memory editOpen MEMORY.md in $EDITOR.
hoangsa-memory memory fact <text...>Append a fact. Tags are comma-separated.
hoangsa-memory memory fact "HTTP retry lives in crates/net/retry.rs"
hoangsa-memory memory fact --tags net,retry "HTTP retry lives in ..."
hoangsa-memory memory lesson --when <trigger> <advice...>Append a lesson.
hoangsa-memory memory lesson \
--when "adding a retry to an HTTP call" \
Use the existing RetryPolicy in crates/net/retry.rs.
hoangsa-memory memory pendingList entries staged in MEMORY.pending.md / LESSONS.pending.md
(only populated when memory_mode = "review" or on lesson conflicts).
hoangsa-memory memory promote <kind> <index>Accept a staged entry. kind is fact or lesson; index is 0-based
from hoangsa-memory memory pending.
hoangsa-memory memory promote lesson 2
hoangsa-memory memory reject <kind> <index> [--reason ...]Drop a staged entry without promoting.
hoangsa-memory memory reject fact 0 --reason "duplicate of existing fact"
hoangsa-memory memory forgetRun the TTL / capacity sweep. Quarantines lessons whose failure ratio
exceeds quarantine_failure_ratio.
hoangsa-memory memory log [--limit N]Tail memory-history.jsonl — the audit trail of every stage /
promote / reject / quarantine / propose event.
hoangsa-memory memory nudge [--window N]Mode::Full only. Asks the synthesizer to propose new lessons from recent episodes.
hoangsa-memory skills listEnumerate installed skills (under .claude/skills/).
hoangsa-memory skills install [PATH]Without PATH: (re)installs the bundled skills (memory-discipline,
memory-reflect, memory-guide, memory-exploring, memory-debugging,
memory-impact-analysis, memory-refactoring, memory-cli).
With a PATH pointing at a <slug>.draft/ directory (produced by
the agent's memory_skill_propose MCP tool): promotes the draft into
a live skill and removes the draft.
hoangsa-memory skills install # bundled
hoangsa-memory skills install .hoangsa/memory/skills/my-skill.draft # promote draft
hoangsa-memory skills install --scope user # ~/.claude/skills/
hoangsa-memory eval --gold <file>Run precision@k over a gold query set (TOML). Reports P@k, MRR, and per-query latency.
hoangsa-memory eval --gold eval/gold.toml
hoangsa-memory eval --gold eval/gold.toml --mode full -k 16
hoangsa-memory eval --gold eval/gold.toml --mode both # side-by-side Zero vs Full
--mode full / both requires --embedder and/or --synth, plus a
stopped daemon (the redb lock is exclusive).
hoangsa-memory domain sync --source <adapter>Pull business rules from an external source (file, notion,
asana, …) into <root>/domain/<context>/_remote/<source>/. See
hoangsa-memory domain sync --help for per-adapter flags.
--root PATH — defaults to ./.hoangsa/memory. Point at ~/.hoangsa/memory for
user-global memory.--json — machine-readable output (for subcommands that support it).--embedder <voyage|openai|cohere> — Mode::Full semantic search.--synth <anthropic|…> — Mode::Full LLM synthesizer.-v / -vv / -vvv — tracing verbosity.npx claudepluginhub unknown-studio-dev/hoangsa --plugin hoangsaThis skill should be used before any non-trivial coding action — editing code, writing new files, running migrations, deploying, or answering a question that involves factual claims about this codebase. It forces the agent to consult hoangsa-memory's persistent memory (USER.md, MEMORY.md, LESSONS.md, the indexed code graph) and acknowledge relevant lessons before acting. Trigger phrases: "edit", "refactor", "implement", "fix the bug in", "add a feature", "deploy", "why does this do", "how does X work".
Checks memory before code exploration when answering questions about code, architecture, or patterns. Saves time by recalling past knowledge.
Organizes, extracts, prunes, and verifies Claude Code persistent memory files to keep MEMORY.md under the 200-line truncation limit and topic files up to date with project state.