🦛 Hippo Memory Insights — 2026-07-08

[github-actions[bot]]

Summary

• Memories in store: 339
• New lessons learned from commits today: 0
• Highest-confidence memory: \ — stale workflow \ files repeatedly cause CI churn; run \ after markdown workflow edits.

Top Memories Surfaced

• \ — stale workflow \ files cause churn and CI friction; run \ after markdown workflow edits.
• \ — use \ as the fast pre-PR validation gate.
• \ — run \ after Go changes and \ after workflow markdown edits.
• \ — \ validates YAML for ANSI escape sequences before other jobs.
• \ — compiler sanitizes ANSI codes in workflow YAML generation via .
• \ — validate workflow changes by compiling a representative workflow with .
• \ — prefer selective Go test patterns such as .

Suggested Improvements

Error Patterns

• Workflow markdown edits still create recurring CI friction when \ files are left stale (, ). Prevent this by tightening authoring guidance around \ edits and ensuring \ is part of every workflow-edit playbook.
• ANSI contamination remains a recurring concern in workflow YAML (, , ). The preventive move is to keep generation paths sanitized and add focused tests around \ and any code that emits workflow YAML.

Code Quality

• The memory store repeatedly surfaces command-development checklist items and semantic type guidance, suggesting uneven enforcement of CLI patterns (, , ). Files under \ and shared workflow typing in \ remain strong candidates for targeted cleanup.
• Deprecated workflow expression forms still appear in memory (). Cleaning remaining uses of \ in \ would reduce warning noise and future migration risk.

CI Health

• Validation discipline dominates the strongest CI memories: run \Syncing actions-lock.json from .github/aw to pkg/actionpins/data/action_pins.json and pkg/workflow/data/action_pins.json...
  ✓ Action pins synced successfully
  Syncing install-gh-aw.sh to actions/setup-cli/install.sh...
  ✓ Action scripts synced successfully
  go build -ldflags "-s -w -X main.version=c4fc1f2" -o gh-aw ./cmd/gh-aw early, then \ before shipping (, , ). This indicates CI failures are often preventable with better local gates rather than new infrastructure.
• Developers are still encouraged to use broad test suites too late or too often; memories favor selective tests and representative compile checks first (, , ). Tightening docs and scripts around targeted validation would reduce wasted cycle time.

Quick Wins

• Add a small validation/helper path that detects modified \ files and reminds or fails fast unless \ has been run recently. Memory: , .
• Add focused tests for ANSI stripping in workflow generation around \ and CI expectations in . Memory: , , .
• Audit and replace remaining deprecated \ references in workflow markdown with . Memory: .

Longer-term Themes

• The repository relies heavily on remembered process rules instead of hard guardrails. Repeated memories about formatting, recompilation, and validation suggest these conventions should be encoded more directly in tooling.
• Workflow authoring remains a high-friction area: YAML hygiene, generated lock files, expression migrations, and compile-time validation appear repeatedly across memories, implying a need for a dedicated workflow-authoring UX pass.

Memory Health

• 
  Hippo - biologically-inspired memory system for AI agents

Usage: hippo (command) [options]

Commands:
init Create .hippo/ structure in current directory
--scan [dir] Find all git repos under dir (default: ~) and init each
--days (n) Days of git history to seed (default: 365 for --scan, 30 for single)
--global Init the global store ($HIPPO_HOME or ~/.hippo/)
--no-hooks Skip auto-detecting and installing agent hooks
--no-schedule Skip auto-creating the machine-level daily runner
--no-learn Skip seeding memories from git history
remember (text) Store a memory
--tag (tag) Add a tag (repeatable)
--error Tag as error (boosts retention)
--pin Pin memory (never decays)
--verified Set confidence: verified (default)
--observed Set confidence: observed
--inferred Set confidence: inferred
--global Store in global store ($HIPPO_HOME or ~/.hippo/)
recall (query) Search and retrieve memories (local + global)
--budget (n) Token budget (default: 4000)
--min-results (n) Minimum results regardless of budget (default: 1)
--json Output as JSON
--why Show match reasons and source annotations
--hops (n) E3.2 multi-hop graph recall: also surface memories
reached by walking the entities/relations graph (n)
hops (0..3, default off) out from the lexical seeds.
Graph hits are tagged [graph: Nhop (rel)]. Today the
graph holds supersedes edges (E3.1); cross-object edges
light up the same traversal once extracted.
--max-neighbors (n) Per-hop fanout cap for --hops (1..200, default 25).
--graph-stream L1: fuse a graph-retrieval stream into RRF, re-ranking
in-pool results by graph proximity to the strong lexical
seeds. Implies rrf scoring (default is blend). Local store
only. Distinct from --hops (which injects out-of-pool
neighbours); this re-ranks within the candidate pool.
--graph-hops (n) Hops for --graph-stream (1..3, default 2).
--graph-seeds (n) Lexical anchors for --graph-stream (default 10). The stream
re-ranks the rank>seeds tail; on a pool with <= n candidates
every candidate is a seed and the stream is inert.
--no-mmr Disable MMR diversity re-ranking
--mmr-lambda (f) MMR balance 0..1 (default: 0.7, 1.0 = pure relevance)
--evc-adaptive ACC-style: when top-K shows high inter-item overlap
(= conflict cluster), expand pool and re-rank by
recency. Default off. RESEARCH.md §PFC.ACC.
--filter-conflicts vlPFC interference filter: drop superseded entries
and 0.3x-downweight entries flagged in an open
conflict with a peer in the same result set.
Uses recorded supersession + conflicts only — never
lexical inference. Default off. RESEARCH.md §PFC.vlPFC.
--value-aware vmPFC value attribution: boost memories with positive
cumulative outcomes and demote those with negative
outcomes during ranking. Multiplier
clip(1 + 0.3*tanh(pos - neg), 0.7, 1.3). Reuses
outcome_positive / outcome_negative; no schema
change. Default off. RESEARCH.md §PFC.vmPFC.
--rerank-utility OFC option-value re-ranker: combine relevance,
strength, and integration cost into a single utility
= score * (0.5 + 0.5 * strength) * (1 - cost_factor)
where cost_factor = min(0.3, tokens / 10000). Re-sorts
results by utility. Default off. RESEARCH.md §PFC.OFC.
--reranker (name) Apply a reranker pass after retrieval
(cross-encoder|llm). Looks up the named
reranker from src/rerankers/index.ts and re-orders
the top-K candidates. Default unset (no reranker).
See docs/plans/2026-05-10-f6-reranker-hardening.md.
--reranker-top-k (n) Cap candidates passed to the reranker (default 50).
--goal (tag) dlPFC goal-conditioned recall: memories tagged with
the goal tag get a 1.5x score boost and results are
re-sorted. Default off. RESEARCH.md §PFC.dlPFC.
--session-id (id) Session identifier for dlPFC goal-stack boost.
Defaults to $HIPPO_SESSION_ID. When set and the
(tenant, session) has active goals (see
'hippo goal push'), recall auto-boosts memories
whose tags match an active goal name. Boost stacks
on top of base BM25 score, capped at 3.0x.
--salience-threshold (n)
Pineal salience: down-weight memories whose
retrieval_count is below n. score *= max(0.5,
retrieval_count / n) for entries with count < n;
entries at or above n are unchanged. Salience emerges
from USE, not from lexical overlap. Default off.
RESEARCH.md §"AI Pineal Gland". (v1's creation-time
lexical gate destroyed LoCoMo 0.28 -> 0.02; this v2
is retrieval-side, opt-in only — see MEMORY.md
"Hippo salience gate destroys benchmark recall".)
--continuity Include continuity block (active task snapshot,
latest matching session handoff, last 5 session
events) above the memory list. Useful at agent
boot when you want both relevant memories AND
where you left off in one call. Anchored on the
active snapshot's session_id; no anchor = no
handoff/events (use 'hippo session resume' for
the explicit handoff-without-snapshot path).
explain (query) Show full score breakdown for each retrieved memory
--budget (n) Token budget (default: 4000)
--limit (n) Cap the number of results displayed
--json Output as JSON
--physics | --classic Force search mode (default: from config)
--no-mmr Disable MMR diversity re-ranking
--mmr-lambda (f) MMR balance 0..1 (default: 0.7, 1.0 = pure relevance)
trace (id) Memory dossier: content, decay trajectory, retrievals,
outcomes, consolidation parents, open conflicts
--json Output as JSON
refine Rewrite consolidated semantic memories with Claude
--limit (n) Cap the number of memories processed this run
--all Ignore llm-refined tag and re-refine everything
--dry-run Call the API but don't write results back
--model (id) Override the default model (claude-sonnet-4-6)
--json Output summary as JSON
(requires ANTHROPIC_API_KEY in env)
eval [(corpus.json)] Measure recall quality against a test corpus
--bootstrap Generate a synthetic corpus from current memories
--out (path) With --bootstrap, write to file instead of stdout
--max-cases (n) With --bootstrap, cap case count (default: 50)
--show-cases Print per-case details (query, R@10, missed, top 3)
--compare (path) JSON from a prior eval --json run; print deltas
--no-mmr Disable MMR for this eval run
--mmr-lambda (f) Override MMR lambda for this run
--embedding-weight (f) Override cosine weight (default: 0.6)
--local-bump (f) Local-over-global priority multiplier (default: 1.2)
--equal-sources Shortcut for --local-bump 1.0
--min-mrr (f) Exit non-zero if mean MRR falls below this
--json Output full summary as JSON
context Smart context injection for AI agents
--auto Auto-detect task from git state
--budget (n) Token budget (default: 1500)
--pinned-only Only inject pinned memories (used by UserPromptSubmit hook)
--include-recent (n) With --pinned-only, also inject the last N writes regardless of pinning
--format (fmt) Output format: markdown (default), json, or additional-context (Claude Code hook JSON)
--framing (mode) Framing: observe (default), suggest, assert
sleep Run consolidation pass (auto-learns + dedup + auto-shares)
--dry-run Preview without writing
--no-learn Skip auto git-learn before consolidation
--no-share Skip auto-sharing to global store
daily-runner Sweep registered workspaces and run daily learn+sleep
dedup Remove duplicate memories (keeps stronger copy)
--dry-run Preview without removing
--threshold (n) Overlap threshold 0-1 (default: 0.7)
status Show memory health stats
audit [--fix] Check memory quality (--fix removes junk)
github GitHub connector subcommands (backfill, dlq)
backfill --repo (owner/name) [--since ISO] [--max (N)]
Paginated backfill of issues + comments
dlq list List DLQ entries for the active tenant
dlq replay (id) [--force]
Re-ingest a DLQ entry (--force skips sig check)
provenance Provenance coverage gate for kind='raw' rows
--json Output as JSON
--strict Exit non-zero when coverage < 100%
drill Walk down a DAG level-2 summary to its children
--limit N Cap children list (default 50)
--budget N Cap total child token cost (≈ chars/4)
--json Output as JSON
assemble --session (id) Build a session's chronological context window
--budget N Token budget (default 4000)
--fresh-tail N Recent rows always kept verbatim (default 10)
--no-summarize-older Disable older-row summary substitution
--scope Restrict to exact scope (default: deny :private:)
--json Output as JSON
correction-latency Wall-clock lag from receipt to supersession (p50/p95/max)
--json Output as JSON
outcome Apply feedback to last recall
--good Memories were helpful
--bad Memories were irrelevant
--id (id) Target a specific memory
conflicts List detected open memory conflicts
--status (status) Filter by status (default: open)
--json Output as JSON
resolve (conflict_id) Resolve a memory conflict
--keep (memory_id) Memory to keep (required)
--forget Delete the losing memory (default: halve half-life)
snapshot _{Persist or inspect the current active task
snapshot save Save active task state
--task (task)
--summary}

--next-step (step)
--source (source) Optional source label
--session (id) Link snapshot to a session trail
snapshot show Show the active task snapshot
--json Output as JSON
snapshot clear Clear the active task snapshot
--status (status) Mark final status (default: cleared)
session Append or inspect short-term session history
session log Append a structured session event
--id (session-id)
--content (text)
--type (type) Event type (default: note)
--task (task) Optional task label
--source (source) Optional source label
session show Show recent events for a session or task
--id (session-id)
--task (task)
--limit (n) Event limit (default: 8)
--json Output as JSON
session latest Show latest task snapshot + events
--id (session-id) Filter by session
--json Output as JSON
session resume Re-inject latest handoff as context output
--id (session-id) Filter by session
handoff Manage session handoffs for continuity
handoff create Create a new session handoff
--summary (text) Handoff summary (required)
--next (text) Next action for successor
--session (id) Session ID (auto-generated if omitted)
--task (id) Associated task ID
--artifact (path) Related file path (repeatable)
handoff latest Show the most recent handoff
--session (id) Filter by session
--json Output as JSON
handoff show (id) Show a specific handoff by ID
current Show compact current state for agent injection
current show Active task + recent session events (default)
--json Output as JSON
forget (id) Force remove a memory
--archive Archive a raw (append-only) memory instead of deleting
--reason "(why)" Reason recorded on the archive (required with --archive)
inspect (id) Show full memory detail
embed Embed all memories for semantic search
--status Show embedding coverage
watch "(command)" Run command, auto-learn from failures
learn Learn lessons from repository history
--git Scan recent git commits for lessons
--days (n) Scan this many days back (default: 7)
--repos (paths) Comma-separated repo paths to scan
promote (id) Copy a local memory to the global store
share (id) Share a memory with attribution + transfer scoring
--force Share even if transfer score is low
--auto Auto-share all high-transfer-score memories
--dry-run Preview what would be shared
--min-score (n) Minimum transfer score (default: 0.6)
peers List projects contributing to global store
sync Pull global memories into local project
import Import memories from other AI tools
--chatgpt (path) Import from ChatGPT memory export (JSON or txt)
--claude (path) Import from CLAUDE.md or Claude memory.json
--cursor (path) Import from .cursorrules or .cursor/rules
--file (path) Import from any markdown or text file
--markdown (path) Import from structured MEMORY.md / AGENTS.md
--vault (path) Import a markdown-vault FOLDER as kind='raw' notes
(Obsidian/Foam/Dendron). Requires --name (vault).
[--scope (scope)]
--dry-run Preview without writing
--global Write to global store ($HIPPO_HOME or ~/.hippo/)
--tag (tag) Add extra tag (repeatable)
export [file] Export all memories (default: stdout)
--format (fmt) Output format: json (default) or markdown
capture Extract memories from conversation text
--stdin Read from piped input
--file (path) Read from a file
--last-session Read from the most recent agent session transcript
--transcript (path) Explicit transcript path (implies --last-session)
--log-file (path) Tee output to a log file (paired with 'hippo last-sleep')
--dry-run Preview without writing
--global Write to global store ($HIPPO_HOME or ~/.hippo/)
setup One-shot: detect installed AI tools and install all
available SessionEnd+SessionStart hooks
--all Install for every JSON-hook tool, even if not detected
--dry-run Show what would be installed without writing
--no-schedule Skip installing or repairing the daily runner
last-sleep Print the last 'hippo sleep --log-file' output and clear it
--path

Log path (default: ~/.hippo/logs/last-sleep.log)
--keep Print without clearing
codex-run [-- ...args] Launch real Codex behind Hippo's session-end wrapper
hook [target] Manage framework integrations
hook list Show available hooks
hook install (target) Install hook (claude-code|codex|cursor|openclaw|opencode|pi)
claude-code/opencode install SessionEnd+SessionStart;
codex wraps the detected launcher in place
hook uninstall (target) Remove hook
decide "(decision)" Record a decision (first-class object + memory mirror)
--context "(why)" Why this decision was made
--supersedes (mem-id) Supersede the decision backed by this memory id
decide list [--status active|superseded|closed|all] [--limit N]
List decisions (table is authoritative, survives decay)
decide get (id) Show a decision by its table id
decide close (id) Retire (close) an active decision by its table id
incident "(incident)" Record an incident (first-class object + memory mirror)
--context "

Details

" What happened / surrounding detail
--link (mem-id) Link a memory as evidence (repeatable)
incident list [--status open|resolved|closed|all] [--limit N]
List incidents (table is authoritative, survives decay)
incident get (id) Show an incident by its table id
incident resolve (id) Resolve an open incident (open -> resolved)
--resolution "(text)" How it was resolved (required)
incident close (id) Retire (close) an open or resolved incident by its table id
process new "(name)" Record a process map (first-class object + memory mirror)
--step "(text)" An ordered step (repeatable)
--description "(text)" Optional summary of the process
process list [--status active|superseded|closed|all] [--limit N]
List processes (table is authoritative, survives decay)
process get (id) Show a process (with its steps) by its table id
process supersede (id) Record a new version that supersedes an active process
--step "(text)" A step of the new version (repeatable, required)
--change "

" What changed in this version (the delta note)
--description "(text)" Optional summary of the new version
process close (id) Retire (close) an active process by its table id
policy new "(name)" Record a policy (bi-temporal first-class object + mirror)
--text "(rule)" The policy rule/statement (required)
--from "(iso)" Effective-from date (default: now)
--to "(iso)" Effective-to date (optional; open-ended if omitted)
policy list [--status active|superseded|closed|all] [--limit N]
List policies (table is authoritative, survives decay)
policy get (id) Show a policy by its table id
policy asof "(iso-date)" Show active policies in force at a valid-time
--name "(policy)" Filter to one policy by name
policy supersede (id) Record a new version that supersedes an active policy
--text "(rule)" The new rule (required)
--from "(iso)" New effective-from (default: now)
--to "(iso)" New effective-to (optional)
--change "

" What changed in this version (the delta note)
policy close (id) Retire (close) an active policy by its table id
skill new "(name)" Record a skill (reusable agent-followable capability)
--instructions "(txt)" The skill body (required)
--trigger "(when)" Optional: when to apply this skill
skill list [--status active|superseded|closed|all] [--limit N]
List skills (table is authoritative, survives decay)
skill get (id) Show a skill by its table id
skill export Render active skills as an AGENTS.md/CLAUDE.md markdown block
skill supersede (id) Record a new version that supersedes an active skill
--instructions "(txt)" The new skill body (required)
--trigger "(when)" Optional new trigger
--change "

" What changed in this version (the delta note)
skill close (id) Retire (close) an active skill by its table id
brief new "(repo)" Record a repo-scoped project brief
--summary "(text)" The brief body (required)
brief list [--status active|superseded|closed|all] [--repo "(repo)"] [--limit N]
List project briefs (table is authoritative, survives decay)
brief get (id) Show a project brief by its table id
brief supersede (id) Record a new version that supersedes an active brief
--summary "(text)" The new brief body (required)
--change "

" What changed in this version (the delta note)
brief close (id) Retire (close) an active project brief by its table id
brief refresh "(repo)" Auto-assemble the brief from the repo's receipts (path:(repo))
--dry-run Print the assembled brief without writing it
note new "(customer)" Record a customer/account-scoped note
--text "(note)" The note body (required)
note list [--status active|superseded|closed|all] [--customer "(id)"] [--limit N]
List customer notes (table is authoritative, survives decay)
note get (id) Show a customer note by its table id
note supersede (id) Record a new version that supersedes an active note
--text "(note)" The new note body (required)
--change "

" What changed in this version (the delta note)
note close (id) Retire (close) an active customer note by its table id
graph extract Rebuild the entity/relation graph from consolidated objects
(decisions/policies/customer-notes/project-briefs); idempotent
invalidate "(pattern)" Actively weaken memories matching an old pattern
--reason "(why)" Optional: what replaced it
wm Working memory — bounded buffer for current state
wm push Push a working memory entry
--scope (scope) Scope name (default: default)
--content (text) Content to store (required)
--importance (n) Priority 0-1 (default: 0.5)
--session (id) Session ID
--task (id) Task ID
wm read Read working memory entries
--scope (scope) Filter by scope
--session (id) Filter by session
--limit (n) Max entries (default: 20)
--json Output as JSON
wm clear Clear working memory entries
--scope (scope) Filter by scope
--session (id) Filter by session
wm flush Flush working memory (session end)
--scope (scope) Filter by scope
--session (id) Filter by session
dashboard Open web dashboard for memory health
--port (n) Port to serve on (default: 3333)
mcp Start MCP server (stdio transport)
goal dlPFC goal stack (B3) — scoped per session
goal push (name) Push a new active goal; prints the new goal id
--policy (type) schema-fit-biased | error-prioritized |
recency-first | hybrid
--success "(cond)" Optional success condition text
--level (n) Goal level (default: 0)
--parent (goalId) Parent goal id (for sub-goals)
--session-id Override session (defaults to HIPPO_SESSION_ID)
--tenant-id (t) Override tenant (defaults to HIPPO_TENANT)
goal list Show active goals as a table
--all Include suspended/completed goals
goal complete (id) Mark a goal completed
--outcome <0..1> Outcome score; >=0.7 boosts, <0.3 decays recalled mems
--no-propagate Close the goal without applying strength side-effects
goal suspend (id) Move an active goal to suspended
goal resume (id) Move a suspended goal back to active (depth-capped)
auth Manage API keys (A5 stub auth)
auth create Mint a new API key (plaintext shown ONCE)
--label Optional human label
--role (r) admin | member (default: admin; member blocked from /v1/sleep)
--tenant (id) Override tenant (defaults to HIPPO_TENANT)
--json Output as JSON
--global Operate on the global store
auth list List API keys (active by default)
--all Include revoked keys
--json Output as JSON
--global Operate on the global store
auth revoke (key_id) Revoke an API key (subsequent validate fails)
--json Output as JSON
--global Operate on the global store
audit Query the append-only audit log (A5 stub auth)
audit list List audit events for the active tenant
--op (op) Filter by op (remember | recall | promote |
supersede | forget | archive_raw | auth_revoke)
--since (iso) Lower bound on ts (ISO timestamp)
--limit (n) Max events (default: 100, max: 10000)
--json Output as JSON
--global Operate on the global store

Examples:
hippo init
hippo remember "FRED cache can silently drop series" --tag error
hippo recall "data pipeline issues" --budget 2000
hippo context --auto --budget 1500
hippo conflicts
hippo session log --id sess_123 --task "Ship feature" --type progress --content "Build is green, next step is docs"
hippo session latest --json
hippo session resume
hippo snapshot save --task "Ship feature" --summary "Tests are green" --next-step "Open the PR" --session sess_123
hippo handoff create --summary "PR is open, tests green" --next "Merge after review" --session sess_123 --artifact src/foo.ts
hippo embed --status
hippo watch "npm run build"
hippo learn --git --days 30
hippo promote mem_abc123
hippo sync
hippo setup
hippo hook install claude-code
hippo decide "Use PostgreSQL for new services" --context "JSONB support"
hippo incident "Prod outage: DB connection pool exhausted" --context "spike at 14:00"
hippo process new "Release" --step "run tests" --step "bump version" --step "publish"
hippo policy new "Data retention" --text "Delete logs after 90 days" --from 2026-01-01
hippo policy asof 2026-03-01 --name "Data retention"
hippo skill new "Run tests" --instructions "npm test before every commit" --trigger "before commit"
hippo skill export
hippo brief new "hippo" --summary "Agent-memory library; E2 first-class objects in progress"
hippo brief refresh "hippo"
hippo note new "Acme Corp" --text "Renewal call: wants SSO before Q3; champion is the VP Eng"
hippo note list --customer "Acme Corp" --status active
hippo graph extract
hippo invalidate "REST API" --reason "migrated to GraphQL"
hippo export memories.json
hippo export --format markdown memories.md
hippo sleep --dry-run
hippo outcome --good
hippo status reported 339 active memories, 13 episodic merges, 5 new semantic memories, and 5 duplicates removed.

• The consolidation pass flagged 4 low-quality memories worth review with .
• Some recall commands hit , so recall/export orchestration should avoid parallel hippo access in future runs.

Operational notes

• \ found no fix/revert/bug commits in the last 7 days.
• \ auto-shared 255 high-value memories to the global store.
• \ reported 339/339 memories embedded and no pending items.

Next Actions

• Review the 4 low-quality memories with .
• Convert the 3 quick wins above into tracked follow-up work.
• Update the hippo runner to serialize recall/export steps to avoid SQLite lock contention.

[!WARNING]

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• proxy.golang.org

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "proxy.golang.org"

See Network Configuration for more information.

Generated by 🦛 Daily Hippo Learn · 20.2 AIC · ⌖ 21.2 AIC · ⊞ 5.4K · ◷

• expires on Jul 10, 2026, 11:10 PM UTC-08:00