Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add 0xmariowu/Autosearch --skill experience-compactgit clone --depth 1 https://github.com/0xmariowu/AutosearchWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/0xmariowu/autosearch/experience-compact)<a href="https://agentmods.dev/skills/0xmariowu/autosearch/experience-compact"><img src="https://agentmods.dev/badge/skills/0xmariowu/autosearch/experience-compact/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/0xmariowu/autosearch/experience-compact"><img src="https://agentmods.dev/badge/skills/0xmariowu/autosearch/experience-compact.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00069 | $0.01424 |
| Opus 5.5 | $0.00028 | $0.00570 |
| Sonnet 5.5 | $0.00014 | $0.00285 |
| Haiku 4.5 | $0.00007 | $0.00142 |
Grade A, and why
autosearch:experience-compact scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 132 lines — stays where its author put it; the contents beside it link to each section on GitHub.
experience-compact — Per-Skill Digest Updater
Reads a skill's experience/patterns.jsonl + existing experience.md, promotes recurring patterns to the digest, prunes stale rules, and rotates raw events if the file is getting large.
Trigger Conditions
Run compact when ANY of:
- ≥10 new events added to
patterns.jsonlsince last compact. patterns.jsonl> 64 KB.- Last event carried explicit user feedback (
user_feedback: acceptedorrejected). - Session ending and this skill was used at least once during the session.
Runtime AI (or a cron on long-running instances) calls this skill per trigger.
Promotion Rules
A candidate rule is extracted from patterns.jsonl when the same winning_pattern string or good_query template recurs. To promote it into the Active Rules section of experience.md:
seen >= 3(appeared in at least 3 distinct events)success >= 2(at least 2 of those events haveoutcome: success)last_verified <= 30 days(most recent appearance within the last 30 days)
Each promoted rule carries a metadata footer:
seen=5, success=4, last_verified=2026-04-21
Anti-Pollution Rules
- Single success never promotes. Even if an event looks great, one instance is noise — requires 3 occurrences.
- User-rejected / rubric-failed / low-relevance events go only into Failure Modes, never Active Rules.
- User corrections (explicit "no, that's wrong") always promote to Failure Modes with high weight.
- Cold rules expire. Any Active Rule with
last_verified > 30 daysmoves to a "pending revalidation" bucket; if not re-verified in another 30 days, removed entirely.
experience.md Template
# <skill-name> experience
## Active Rules # ≤ 20 entries
- <rule> — seen=5, success=4, last_verified=2026-04-21
- <rule> — seen=3, success=3, last_verified=2026-04-19
## Failure Modes # ≤ 15 entries
- <failure pattern> — seen=4, last_verified=2026-04-20
- <user correction> — seen=1, last_verified=2026-04-18
## Good Query Patterns # ≤ 20 templates
- `{brand} {feature} 翻车 {year}` — seen=7
- `{category} 避雷 小红书 最近` — seen=5
## Last Compacted
- 2026-04-22, from 37 events, promoted 3 rules, archived raw events before 2026-03-22.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 132 lines · 69 tokens per session scan A be5d11ca0fb2
autosearch:experience-compact is a skill published in the GitHub repository 0xmariowu/Autosearch (52 stars, last pushed 2d ago), licensed MIT. It adds 69 tokens to every session and 1,424 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
quarry-capture
Persist something worth keeping into quarry so it survives context compaction and is findable by quarry-recall later — a durable fact, a distilled lesson, or a URL to index. Trigger phrases: "remember this", "save this fact", "note that we decided X", "learn this lesson", "the rule going forward is X", "ingest this…
quarry-recall
Retrieve what quarry already knows before researching from scratch or guessing. Covers three overlapping intents, one tool: understand code ("how does X work", "where is X defined", "what calls X"); recall project knowledge or a decision ("what did we decide about X", "why is it this way", "did we consider Y…
stats
Show Captain Memo's corpus statistics (chunks per channel, observation counts, indexing progress, embedder info). Use when the user types /captain-memo:stats.
search
Hybrid search across Captain Memo's local memory + skills + observations. Use when the user types /captain-memo:search to retrieve top hits without the model having to decide whether to call searchall on its own.
doctor
Health probe across the worker, embedder, summarizer, queue, vector index, schema, cross-AI wiring and plugin install (cache and live versions). Use when the user types /captain-memo:doctor or when something seems wrong.
observations
List recent captured session observations (the summarized voyage logs). Use when the user wants to see what Captain Memo has logged from past sessions.