manage-context
Skill10 starsChanged 19 days agoMITView on GitHub
Show the role every skill plays in a project (Auto, name-only, /slash only, off; installed globally, per project or by a plugin; switched differently in other projects) and switch skills, plugins, MCP servers and built-in tools on or off. Measures what fills a session's starting context, mines 90 days of usage for recommendations, and applies changes with a dry run and undo.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- KMaruthi2002/universal-llm-skillstheir own version, first seen Oct 25, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- KMaruthi2002/universal-llm-skills★ 1same 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 repository25 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agents-md-improverSkill
- ai-tutorialSkill
- deassembleSkill
- excalidraw-diagramSkill
- find-overlapping-skillsSkill
- first-principlesSkill
- humanizerSkill
- improve-skillsSkill
- local-artifactSkill
- loom-transcriptSkill
- organize-skill-mcpSkill
- pair-consultSkill
- pair-optimizeSkill
- pair-ratchetSkill
- pr-classifySkill
- pr-reportSkill
- scan-sessions-and-learnSkill
- simplify-pythonSkill
- skill-flowSkill
- slack-commSkill
- socratesSkill
- step-by-stepSkill
- sub-consultSkill
- to-dbmlSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 61
Leave --out off when there's no scratchpad.
- Very specific: line 102
Never print values from ~/.claude.json.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Manage context
Users come with one of three asks. Pick the entry point by the ask:
| Ask | Run |
| "what does each skill do here?", "which are on?" | scripts/skills.py list (terminal), or the full report (§1) when tokens and usage matter |
| "turn X off", "make Y slash-only", "enable plugin Z" | scripts/skills.py set (Quick switch) |
| "slim my context", "what uses my tokens?" | the full report, §1–3 |
Roles
| Role | Model sees | /name | Why pick it |
Auto (on) | name + description, may run it unasked | yes | skills the model should reach for itself |
| Name only | name only | yes | rarely needed, but keep it reachable for a few tokens |
/slash only (user-invocable-only) | nothing | yes | workflows you start yourself: zero tokens, never fires by surprise |
| Off | nothing | no | not wanted here |
A skill's description says how it should be used, but config decides. A skill written as "only when the user types /x" is still Auto if no setting says otherwise. The overview exists to expose that gap. Who decides, first match wins:
- plugin skills follow their plugin's on/off, never per-skill settings
- managed policy (locked)
disable-model-invocation: truein the skill's SKILL.md, locked at /slash only (only Off changes it)- the highest settings layer that names it: this project's
settings.local.json> the project's sharedsettings.json> global~/.claude/settings.json - Auto
scripts/inventory.py implements this and was checked against a live session's
listing (0 mismatches). Read references/levers.md before explaining an ignored setting or a rejection.
Quick switch
python3 <skill-dir>/scripts/skills.py set NAME... auto|name|slash|off [--scope global|project]
python3 <skill-dir>/scripts/skills.py set --plugin NAME... on|off [--scope global|project]- Without
--scopeit writes where the current value comes from: this project if a project setting decides the skill here, else global. A global value would lose to the project setting, so the switch would do nothing. Pass the scope if the user named one. - Show the dry run verbatim (role before → after, file edits,
!warnings,✗rejections), get an explicit yes, re-run with--apply, and report the undo command. - Claude Code's own
/skillsmenu also switches skills, but only in this project'ssettings.local.json. This script adds global scope, plugins, bulk changes and undo.
1. Scan (full report)
python3 <skill-dir>/scripts/scan.py --project "$PWD" --out <scratchpad>/context-report- Leave
--outoff when there's no scratchpad; the default is~/.claude/context-reports/<project-slug>/. - It measures the current session (
$CLAUDE_CODE_SESSION_ID) from turn 1, so mid-session is fine;--session <id|path>measures another. If it printsfirst turn: None(skill run as the first message), run it again before reporting. - Token numbers reflect config at that session's start; skill roles reflect config now. The page warns when the two differ. After a change, measure a new session.
--history-days N/--no-historytune usage mining;--mode bullopens the aggressive recommendations (default bear). Match the user's stated appetite.
Open the HTML and reply with: the first-turn total, the 5 largest items, the
skills here: line (count per role), the ignored settings by cause (say what was
meant, why it does nothing, the fix), and the bear/bull lines
with their top items. The page has one module per left tab, each with a one-line
goal: Baseline (opens first: the treemap is the core answer to "where do my
tokens go"), Skills (one switch per skill and plugin), Recommendations,
Ignored settings (one card per cause, with fix buttons), Plan. Every switch feeds the Plan tab: Copy plan and paste it
here. Link a tab with #skills, #recs, #plan on the HTML path.
Quote a recommendation's reason when asked why ("listed in 407 sessions, model never
called it · you typed /name 12× (still works)"); rules are in references/recommendations.md.
For "where/when did I use X?", each Usage cell lists its 3 latest sessions: click copies
claude --resume <id> for that project, ↗ opens the raw transcript.
2. Dry run
Save pasted JSON ("manage_context_plan": 1) to <scratchpad>/context-plan.json
(or use the given path, usually ~/Downloads/context-plan.json), then run
python3 <skill-dir>/scripts/apply_plan.py <plan> --project "$PWD". Show the output
verbatim and explain each ✗ with levers.md. Ask for an explicit yes; approval of an
earlier plan doesn't count.
3. Apply
Re-run with --apply. Report the edit count, the backup directory, the undo command
(apply_plan.py --undo <file> --apply, dry run first too), and that changes take
effect in the next session.
Rules
- Change config only through these scripts. They re-read files right before an atomic write, back up first, and write an undo file.
- Never print values from
~/.claude.json; the scripts read only its keys and mask MCP configs. - Token numbers are calibrated estimates: say "~".
- Recommendations are evidence, not a verdict; the user decides. /slash only is
reversible and keeps
/nameworking. - Usage counts come from tool calls and typed commands only. Never quote message text from other sessions.
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.