memory
Skill136 starsChanged 3 days agoMITView on GitHub
Autonomous user preference detection, storage and injection. The agent detects preferences from conversation, saves them to Postgres, and loads active ones at the start of every flow.
110 other projects use a version of this.
Adopted by
Most recently taken up by
- eduardoweberai-cloud/claude-skillstheir own version, first seen Nov 16, 2026
- arxdsilva/ai-loopstheir own version, first seen Nov 16, 2026
- raucheacho/lorag-claude-codetheir own version, first seen Nov 16, 2026
- matthewwong525/WongStacktheir own version, first seen Nov 15, 2026
- maxwell2732/claudecode-research-harness-workflowtheir own version, first seen Nov 15, 2026
- XORwell/archeflowtheir own version, first seen Nov 15, 2026
Used with
Projects with it also use ecosystem-reference, fractal-delegation and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
naman10parikhSkillecosystem-reference
Quick-access directory of Claude Code ecosystem repos, skills, plugins, MCPs, and patterns. Use when improving the harness.
Also in 5 projects that use this
naman10parikhSkillfractal-delegation
Also in 5 projects that use this
roronoazoroshao369Skillmemory-compression
Also in 5 projects that use this
Maliot100XSkillsend-user-message
Send a proactive message to a paired user via their channel (Discord, Telegram, or WhatsApp). Use when the agent needs to notify, alert, or send an unsolicited message to a user — especially during heartbeat invocations, scheduled tasks, or when the agent wants to reach out without a prior user message in the current conversation. Triggers: 'send message to user', 'notify user', 'alert user', 'message the user on discord/telegram/whatsapp', or any need to proactively communicate with a paired sender.
Also in 4 projects that use this
jh941213Skilldocs-ci
문서 CI(링크·OpenAPI lint·breaking 게이트·mermaid 검증)나 문서 드리프트 감지(docs.yaml 신선도 검사)를 설치하거나 돌릴 때 사용. 문서 내용 작성은 docs-architecture·docs-interfaces·docs-manuals.
Also in 3 projects that use this
jh941213Skilldocs-interfaces
API·이벤트 인터페이스 문서(OpenAPI·swagger·AsyncAPI 명세, 엔드포인트 레퍼런스, API 구성도·시퀀스, API CHANGELOG)를 만들거나 코드와 맞출 때 사용. 시스템 구성도는 docs-architecture, API 설계 원칙은 api-design-principles.
Also in 3 projects that use this
- +5 more from naman10parikh
- +1 more from jh941213
Other versions110 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 110
- google-gemini/gemini-cli★ 109ksame name
- jlia0/tinyclaw★ 3.6ksame name
- TinyAGI/tinyagi★ 3.6ksame name
- TinyAGI/tinyclaw★ 3.6ksame name
- Chachamaru127/claude-code-harness★ 3.2ksame name
- codeaholicguy/ai-devkit★ 1.7ksame name
- Muuuun/luxas★ 1.2ksame name
- shobcoder/shob★ 575same name
- boshu2/agentops★ 454same name
- makushingu/harness-maxxing★ 235same name
100 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository15 other files it gives its agents.
- AGENTS.mdAGENTS.md
- acp-connectorSkill
- applySkill
- browser-automationSkill
- dailySkill
- dashboardSkill
- dbSkill
- newsSkill
- onboardingSkill
- polishSkill
- profileSkill
- radarSkill
- referralsSkill
- strategySkill
- targetsSkill
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 5 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 93 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Detection: after every user message, evaluate whether a preference was stated, implied, or corrected
- Distinctive: line 19
After every user message, scan for 3 signal types:
- Distinctive: line 36
Strategy level changes: When the user's situation changes (employment status, urgency), detect and propose a strategy level change.
- Very specific: line 48
All preferences live in the preferences table, accessed via scripts/db.js:
- Very specific: line 86
When a correction is detected (user contradicts a stored preference), the same INSERT with ON CONFLICT DO UPDATE handles it.
- Very specific: line 131
Don't duplicate: If a preference is already in users.data.profile or users.data.style_profile, skip it.
- Very specific: line 132
Single source of truth per (category, key): Use ON CONFLICT DO UPDATE, never insert duplicates
- Distinctive: line 135
Never ask "should I save this?": If it passes the save checklist, save it silently
- Distinctive: line 136
Report what was saved: After saving, briefly mention: "Guardé que preferís X" / "Saved preference: X".
“Common” is measured: kept by at least 6 in 10 of the 93 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Memory — Autonomous preference management
This skill implements Gold Rule 3 (user preferences always up to date). The agent proactively detects, stores, and injects user preferences without being asked.
When it runs
Always. This is not a flow triggered by a keyword. It is a background behavior that runs during every interaction:
- Detection: after every user message, evaluate whether a preference was stated, implied, or corrected
- Injection: at the pre-flight of every flow, load active preferences into context
Detection
After every user message, scan for 3 signal types:
| Signal | Example | Confidence | Source |
| Explicit | "no quiero empresas de crypto" | 1.0 | explicit_statement |
| Implicit | "respondeme corto" | 0.7 | inferred |
| Correction | "actualmente busco roles de IC, no manager" | 1.0 | correction |
Save vs skip checklist
Save (proactively, no need to ask):
- Job search preferences (roles, industries, locations, work mode, salary)
- Communication preferences (language, tone, length, format)
- Compensation criteria (range, equity, benefits)
- Tooling/workflow preferences (which platforms, how to apply)
- Corrections to anything previously stored
- Explicit requests: "recordá que..." / "remember that..."
- Strategy level changes. When the user's situation changes (employment status, urgency), detect and propose a strategy level change. See AGENTS.md "Strategy levels"
Skip:
- Trivial/obvious info ("user asked about Python")
- Already in CV or
users.data.profile(don't duplicate) - Already in
users.data.style_profile - Re-discoverable facts (can web search)
- Session-specific ephemera (temporary file paths, one-off debugging)
- Already in AGENTS.md or other context files
Storage
All preferences live in the preferences table, accessed via scripts/db.js:
preferences (
id SERIAL PRIMARY KEY,
user_id INTEGER REFERENCES users(id),
category TEXT NOT NULL, -- job_search, communication, compensation, tooling, workflow
key TEXT NOT NULL, -- e.g. "avoid_industries", "reply_language", "salary_min"
value TEXT NOT NULL, -- e.g. "crypto,gambling", "spanish", "5000"
confidence REAL DEFAULT 1.0, -- 1.0 explicit, 0.7 inferred, 0.5 auto-summarized
source TEXT DEFAULT 'explicit_statement', -- explicit_statement | inferred | correction
status TEXT DEFAULT 'active', -- active | superseded
created_at TIMESTAMPTZ DEFAULT NOW(),
updated_at TIMESTAMPTZ DEFAULT NOW(),
UNIQUE(user_id, category, key)
)Categories
| Category | Examples |
job_search | avoid_industries, target_roles, location_preference, work_mode, visa_requirements |
communication | reply_language, reply_tone, reply_length, greeting_style, avoid_bullets |
compensation | salary_min, salary_max, currency, equity_expectation, required_benefits |
tooling | preferred_platforms, apply_method, browser_mode |
workflow | application_batch_size, follow_up_timing, auto_apply_threshold, strategy_level |
Save a new preference
node scripts/db.js "INSERT INTO preferences (user_id, category, key, value, confidence, source) VALUES (<user_id>, '<category>', '<key>', '<value>', <confidence>, '<source>') ON CONFLICT (user_id, category, key) DO UPDATE SET value = EXCLUDED.value, confidence = EXCLUDED.confidence, source = EXCLUDED.source, updated_at = NOW()" --writeThe ON CONFLICT clause handles updates: if the preference already exists, it replaces the value and bumps updated_at. No need to check first.
Correction handling
When a correction is detected (user contradicts a stored preference), the same INSERT with ON CONFLICT DO UPDATE handles it. The old value is replaced, source becomes correction, and updated_at is bumped. The old value is not preserved (single user, no audit trail needed).
Strategy level detection
The agent must detect signals that the user's job search situation has changed and propose a strategy level adjustment. This is Gold Rule 3 applied to urgency/aggressiveness.
| Signal | Example | Proposed level |
| Lost job / fired | "me despidieron", "me quedé sin trabajo", "lost my job" | active |
| About to lose job | "me van a despedir", "termina mi contrato en X", "my contract ends" | active |
| Desperation | "necesito algo ya", "urgentísimo", "need a job now" | aggressive |
| Found a job | "encontré trabajo", "acepté una oferta", "got the job" | passive |
| Employed, casually looking | "estoy viendo opciones", "open to opportunities" | selective |
| Wants more aggressive | "aplica más agresivo", "send more applications" | bump up one level |
| Wants less aggressive | "frená un poco", "too many applications" | bump down one level |
When a strategy change is detected:
- Propose the change to the user (never change without asking)
- Explain what will change (batch sizes, match threshold, relax rules)
- Wait for confirmation
- Save to
preferences(workflow.strategy_level) andusers.data.strategy - Report: "Estrategia actualizada:
<level>"
Injection
At the pre-flight of every flow, load active preferences:
node scripts/db.js "SELECT category, key, value, confidence, source FROM preferences WHERE user_id = <user_id> AND status = 'active' ORDER BY category, key"Inject the result into the flow's context. Treat preferences as constraints:
job_search.*→ filter jobs, discard non-matchingcommunication.*→ shape drafts (language, tone, length)compensation.*→ filter by salary, negotiatetooling.*→ choose platforms and methodstooling.browser_mode→ controls browser visibility in all flows that use the browser (agent-browser). The preference maps toagent-browser.jsonheaded/--headed/--headlessflags. Valid values:headless(always headless except manual login/2FA),headed(always headed),headed_logins_only(headed only for logins/2FA, headless otherwise),ask_each_time(agent asks before each browser session). Default if not set:headed_logins_only. Manual login/2FA is always headed regardless of this preference (Gold Rule 5)workflow.*→ tune batch sizes and timing
Confidence-aware: preferences with confidence < 1.0 (inferred) can be overridden by explicit user instructions in the current session. Preferences with confidence = 1.0 (explicit) hold unless the user explicitly changes them.
Rules
- Proactive, not reactive. Save when detected, don't wait to be asked
- All DB access via
scripts/db.js(seedbskill) - Don't duplicate. If a preference is already in
users.data.profileorusers.data.style_profile, skip it. Thepreferencestable is for things said in conversation that aren't captured by the structured profile - Single source of truth per (category, key). Use
ON CONFLICT DO UPDATE, never insert duplicates - Corrections replace, not append. "actually I want X" updates the existing value
- Inferred preferences are weaker. Mark
confidence = 0.7andsource = 'inferred'. They can be overridden by explicit statements - Never ask "should I save this?". If it passes the save checklist, save it silently
- Report what was saved. After saving, briefly mention: "Guardé que preferís X" / "Saved preference: X". One line, no ceremony
Dependencies
- Depends on
onboarding(DB +userstable must exist) - Uses
dbskill for all DB access - Consumed by every flow via pre-flight injection
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

