remembering-conversations
Skill4 starsChanged 48 days agoMITView on GitHub
Recall past OpenCode conversations when the user references previous work, past decisions, or earlier sessions — "how did we handle X", "the conversation about Y", "what did we decide", "we tried this before", "last week we...". Use episodic_search, then episodic_read_window for a bounded window, and episodic_read_session only when the full session is needed.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- PracticalSwan/agent-skillstheir own version, first seen Oct 19, 2026
- obra/episodic-memorytheir own version, first seen Oct 8, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- obra/episodic-memory★ 487same name
- PracticalSwan/agent-skills★ 22same name
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 repositoryOne other file it gives its agents.
- AGENTS.mdAGENTS.md
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Do NOT search for questions answerable from the current codebase or the current conversation — the index is for cross-session recall, not …
- Distinctive: line 56
If search returns nothing relevant, say "I don't have a past conversation about that" rather than confabulating.
- Very specific: line 61
Only OpenCode sessions are indexed (anything before the OpenCode switch is not, unless it lives in opencode.db).
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Remembering Conversations
You have episodic memory: every past OpenCode session is indexed and searchable via three native tools.
When to search
Search proactively when:
- The user references prior work: "like we did with X", "the conversation about Y", "what did we decide about Z", "we tried that before"
- You're about to propose an approach the user may have already evaluated or rejected
- A bug or error message feels familiar ("didn't we see this before?")
- The user asks about their own history: "when did we set up X", "which repo was Y in"
Do NOT search for questions answerable from the current codebase or the current conversation — the index is for cross-session recall, not code search.
How
episodic_searchwith a natural-language query describing the topic and intent, not exact keywords ("migrating from Claude Code to OpenCode", not "claude opencode").- Narrow with
after/beforedates. Local indexes also support an exacttextsubstring (an error string, a flag name, a file path). - On local indexes, use
mode: "text"for lexical BM25 search: every query word must appear (token-based AND, BM25-ranked) — not phrase/adjacency or substring matching. - Remote indexes support only vector search, without
textfiltering. Include relevant terms inquery; do not retry unsupported text/hybrid modes.
- Narrow with
- Skim the returned excerpts (date, session title, score, and anchor). Vector
similarity scores are NOT calibrated probabilities: ≥ ~0.55 is a strong
match, 0.4–0.55 is likely relevant, and < ~0.35 is weak or merely adjacent.
These thresholds apply only to vector results. For hybrid results, use the
snippet and the
rrflabel instead; RRF scores are on a different scale. Say when the corpus doesn't really contain the topic. episodic_read_windowwith the result's session ID, anchor message ID, andsource_id(required for remote indexes) to inspect a small window first. Current-source windows read privacy-gated live messages. Foreign-source windows read labeled indexed exchanges:before/aftercount chunks, not messages, and content may be stale until that source syncs. Missing or stale anchors cannot expand. Stop on privacy denials rather than trying indexed reads.episodic_read_sessionwith the session ID and the samesource_idwhen the window is insufficient or unavailable. Useindexed: truefor indexed excerpts, including legacy unanchored hits. Do not present these as a full live transcript.
Answering
- Cite what you found with its date and session title ("on 2026-07-19, in 'Fix login User-Agent to get past the bot wall', we concluded...").
- Distinguish "we decided X" from "we tried X and abandoned it" — the transcript usually records the verdict; report it accurately.
- If search returns nothing relevant, say "I don't have a past conversation about that" rather than confabulating.
Limits
- Only OpenCode sessions are indexed (anything before the OpenCode switch is not,
unless it lives in
opencode.db). - Conversations containing the marker
DO NOT INDEX THIS CHATare excluded — that includes conversations about this tool itself that quote the marker. - Excerpts embed user/assistant text and tool names, not tool output.
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.

