inscribe
Skill4 starsChanged 2 days agoApache-2.0View on GitHub
Use when you are Hydra and the owner says /inscribe or asks you to save what you learned, before a context reset or compaction, or when the digest says a memory is nearly full. Routes each fact to owner, project or Hydra memory and curates them; use with a project to save project context.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- Alexander-Tyagunov/magiciantheir own version, first seen Sep 28, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- Alexander-Tyagunov/magician★ 14same 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 repository17 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- audit-testsSkill
- deslopSkill
- follow-conventionsSkill
- review-codeSkill
- write-ci-workflowSkill
- write-front-endSkill
- write-pull-requestSkill
- write-skillSkill
- pnyx-arrange-columnsSkill
- pnyx-cliSkill
- pnyx-firstmate-as-hydraSkill
- pnyx-manage-workersSkill
- pnyx-orchestrateSkill
- pnyx-report-progressSkill
- pnyx-set-upSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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 11
Keep saved context brief and up to date, and edit or remove entries when things change.
- Very specific: line 29
Read the current memory: pnyx-axi owner memory list and pnyx-axi hydra memory list --hydra <hydra> (each shows the characters used.
≈ “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
inscribe
Keep saved context brief and up to date, and edit or remove entries when things change. Save each fact at the narrowest level that fits:
- Owner memory, Pnyx-wide and read by every Hydra (24000 characters): the owner's preferences under
preferences, general learnings about tools, workers and routes underlearnings, and standing worker instructions for every Hydra underworkers(1500 characters, added verbatim to every new worker header; change them withpnyx-axi owner worker-notes set --text "<whole document>" --reason "<why>"and keep the key). - Project memory, shared by every Hydra on that project (10000 characters): facts, conventions and owner preferences that hold only for that project.
- Hydra memory, this Hydra only (1500 characters): its scope, where it left off, its check cursor. Successful inscribe milestones track what has been read.
A fresh session and a session after compaction receive owner memory within a 3000 estimated-token budget and Hydra memory in full. Read pnyx-axi owner memory list when a decision needs omitted owner memory. A resumed session receives only changed digest sections, and later wakes carry a short memory-changed note. Project memory loads only on demand and for workers assigned to that project, so every line must earn its place.
When invoked with a project (/inscribe <project>), resolve it with pnyx-axi project list --hydra <hydra> and read pnyx-axi project memory list --project <name|path> --hydra <hydra>. Route facts, conventions and owner preferences specific to that project into its shared memory. All its documents share a 10000-character budget across Hydras; a nearly-full digest asks you to curate it. General owner preferences and learnings go to owner memory.
profile: build/test/lint commands, PR rules, key skills and when to use them, main conventions. Refresh missing or out-of-date profiles from the default-branch sources listed by Pnyx, reading only changed files. Save with the listed--sources <fingerprint>so a source change cannot be acknowledged accidentally.preferences: the owner's preferences for this project. Keep them until the owner changes them; a profile refresh does not replace them.learnings: durable discoveries about this project, without task progress or history.
Save each document whole with pnyx-axi project memory set --project <name|path> --key <profile|preferences|learnings> --text "<document>" --reason "<why>" --hydra <hydra> (add --sources <fingerprint> for a profile). Use project memory forget with the project and key to remove a document. Over-budget saves are refused: shorten and retry. Replacements and removals stay in the project's append-only log. Finish the project pass with project memory save as described below, then report what changed and the project characters used.
Without a project, sweep for every layer: route project facts and project preferences as above, and curate owner and Hydra memory with the steps below.
- Read the current memory:
pnyx-axi owner memory listandpnyx-axi hydra memory list --hydra <hydra>(each shows the characters used; owner memory also prints its revision). - Start or resume the pass with
pnyx-axi hydra inscribe start --hydra <hydra>. Keep itsmilestoneID andthrough, and run the history command it returns. It reads since the last successful inscribe, including the owner's messages, your replies read from your native transcripts, task briefs, intents and helper results. Helper results are historical context with their task and outcome status; unchanged waits and duplicate automated chat copies are omitted. Follow every--after <cursor>untilmoreis false; each page ends at a whole message. When more follows, carry forward the facts from this page for your next update before finishing the pass. Keep the owner's messages and the briefs' standing rules:- Owner preferences: how the owner wants work done, standing decisions, people and tools they named. These hold until the owner changes them.
- Learnings: facts you had to discover the hard way about tools, workers and routes, which would cost the next session time to rediscover. Each must still be true; check one you doubt before keeping it.
- Hydra notes: what this Hydra covers and where it left off.
- Leave out task progress, conversations, PRs and anything a task's history, brief or status already records. The exception is a standing rule the owner states with "always", "every" or "never": save it as a preference even when a brief also holds it, because the brief ends with its task.
Curate each document: merge duplicates, rewrite a changed preference in place, drop learnings that turned out wrong or are no longer needed, move a Hydra entry that holds for every Hydra into owner memory, and keep one plain line per entry, most important first. Learnings have two kinds. Normal learnings become stale after 30 days without confirmation or 10 memory saves without use, whichever comes first. Short-lived for now learnings become stale after 7 days or 3 unused saves. New lines default to normal. To add for now lines, save the whole learnings document with
--kind for-now --until <condition>; these flags apply only to new lines. The condition is required:task:<hydra>/<task-key-or-id>, a full PR URL,version:<version>, ordate:YYYY-MM-DD(expires at midnight UTC). Pnyx checks task completion and followed PR merges against its saved state, and dates against the review time. Version conditions need your evidence; unknown task or PR state never proves completion.Review every stale or expired line. Confirm only with evidence from this session:
owner memory confirm --text "<exact learning line>" --reason "<session evidence>", orproject memory confirm --project <project> --text "<exact learning line>" --reason "<session evidence>". Confirmation refreshes the date and clears the unused-save count. Archive an unproven, outdated or expired line withowner memory archive --text "<exact learning line>" --reason "<why>"or the project equivalent.memory listshows every line an archive, set or forget removed, with the date and reason of that change, for recovery through a new set. Preferences stay undated and never decay.When owner memory is nearly full, move clearly project-specific learnings into that project's memory yourself before shortening or dropping useful facts. Resolve the attached project, read its memory, then run
pnyx-axi owner memory move --project <name|path> --text "<exact learning line>" --reason "<why this belongs to the project>" --if-revision <owner revision> --hydra <hydra>. The move is one durable step and refuses growth beyond the project budget; curate the project first if necessary. Reload both memories after a move before saving whole documents.Owner preferences are the owner's own words. Before moving any preference out of owner memory, ask the owner through a real
pnyx-axi task ask <task> --to owner --question "<preference and proposed project>"and wait for the answer. Name the preference and destination project; a status update or rhetorical question is not approval. Keep it in owner memory until the owner approves.Save each changed document whole, with a reason:
pnyx-axi owner memory set --key preferences --text "<document>" --reason "<what changed>" --if-revision <revision> pnyx-axi owner memory set --key learnings --text "<document>" --reason "<what changed>" --if-revision <revision> pnyx-axi hydra memory set --key scope --text "<text>" --reason "<what changed>" --hydra <hydra>Before writing
left-off, put each wait or next step on its own task:task depend <task> --on <prerequisite>for another task,task ask <task> --to ownerfor the owner,task brieffor a next step,task createfor untracked work.left-offthen links those tasks as[@Title](pnyx-task:<id>)references and says what comes next; the timed review flags it once they are all done, or after 7 days.Forget keys that became stale with
owner memory forget --key <key>orhydra memory forget --key <key> --hydra <hydra>, each with--reason "<why>". A save over budget is refused: shorten and save again. Other Hydras share owner memory, so a save refused for its revision means another Hydra saved first: list it again, merge and save.- Finish each inscribe pass by saving each memory layer you curated:
pnyx-axi owner memory save --reason "<what you checked>"and/orpnyx-axi project memory save --project <project> --reason "<what you checked>" --hydra <hydra>, even when the documents did not change. Add--used "<exact line>"for a learning you used or confirmed this session; for several, put their exact lines in one newline-separated value. Name the session evidence in the reason. This save refreshes their dates and clears their counts, and adds one unused save to every other learning in that layer. Include lines you already confirmed during this pass in--usedtoo. A set, read, wake or timed review never advances this count. Recheck any lines that reach their limit on this save and confirm or archive them before finishing. - Only after every history page is read and the saves above succeed, finish the milestone with
pnyx-axi hydra inscribe finish --key <id> --through <through> --hydra <hydra>, using the pass snapshot returned in step 2. On failure, recordhydra inscribe fail --key <id> --error "<why>" --hydra <hydra>; it leaves the previous boundary in place. If you are supervising withhydra check, save the lastthroughyou fully handled ashydra memory set --key check-cursor --text "<through>". - During an agent switch, return silently. Otherwise finish with one line: what you saved, what you dropped, and the characters now used in each layer. Forgotten and replaced entries stay in their logs.
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.

