Proc context budget

Skill barcelosvinicius/basic-engineering/plugins/be/skills/proc-context-budget

Claude Code plugin + npm base for AI-assisted engineering: 25 skills, 12 agents, slash commands, session-continuity hook. Also works with Copilot, Cursor, and others.

Install

npx -y skills add barcelosvinicius/basic-engineering --skill proc-context-budget

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when the session feels sluggish, output quality drops, after adding several skills/agents/MCP servers, or to decide whether there is room to add more — audits what consumes the context window (agents, skills, MCP tools, rules, CLAUDE.md) and returns prioritized token-savings recommendations.

SKILL.md

3.0 KB, 690 tokens by cl100k_base, as published. Nobody here has run it

Context Budget

Make the "unused = context noise" principle measurable. Audit token overhead across every always-loaded component and surface the cheapest cuts. The biggest lever is almost always MCP tools, then agent descriptions and CLAUDE.md.

When to use

The session feels slow or output quality is degrading.
You just added several skills, agents, or MCP servers.
You want to know how much headroom is left before adding more.
Backs the /be:context-budget command.

What costs context (and when)

Component	Loaded	Rough cost
MCP tools	always, while a server is enabled	~500 tokens per tool (schema)
Agent `description`	always (every Task invocation)	the frontmatter description, every spawn
CLAUDE.md chain	always	the whole file (project + user)
Always-on rules	always	each file
Skill `description`	always (body only on trigger)	one line per skill; bodies are on-demand
Agent / skill bodies	on demand	free until invoked

Key insight: skills are cheap (progressive disclosure — only the description is always loaded). MCP servers and CLAUDE.md are where bloat hides.

How it works

Inventory — estimate tokens per component (words × 1.3 for prose, chars / 4 for code/JSON). Count MCP servers and total tools (.mcp.json).
Classify each into a bucket:
- Always needed — referenced in CLAUDE.md / backs an active command / matches the project's stack → keep.
- Sometimes — domain/stack-specific, not referenced → activate on demand.
- Rarely — no reference, overlaps another component → remove or lazy-load.
Flag issues: MCP over-subscription (>10 servers, or servers that wrap a CLI you already have like gh/git/npm); bloated agent descriptions (>30 words load on every spawn); CLAUDE.md bloat (>~150 lines / outdated sections); duplicate components.
Report ranked by token savings.

Report format

Context Budget — estimated overhead ~XX,XXX tokens
| Component  | Count | ~Tokens |
| Agents     | N     | X,XXX   |
| Skills     | N     | X,XXX   |
| MCP tools  | N     | XX,XXX  |
| CLAUDE.md  | N     | X,XXX   |

Top savings:
1. <action> → ~X,XXX tokens
2. ...

Best practices

MCP first: a 30-tool server costs more than all your skills combined. Disable what the current task doesn't need; pair with mcp.recommended.json.
Keep CLAUDE.md ≤ ~150 lines — move durable conventions into skills/rules.
Agent descriptions are always loaded — keep them one tight sentence.
Re-audit after adding any agent, skill, or MCP server to catch creep early.

Gives 0 of the 12 instructions most memory context skills give in 690 tokens

Counted across 674 of the 847 authors here whose files we hold, read 2026-08-06

inform the user when setup is completein 21 of 674, across 6 files
confirm the draft with the user before writingin 21 of 674, across 6 files
update the agent skills block in place if it existsin 21 of 674, across 6 files
present findings to the userin 20 of 674, across 5 files
write the three docs files from seed templatesin 20 of 674, across 5 files
ask the user about each decision one at a timein 19 of 674, across 4 files
edit CLAUDE.md if it existsin 18 of 674, across 3 files
explore current repo statein 18 of 674, across 3 files
do not overwrite user edits to surrounding sectionsin 18 of 674, across 3 files
back up the original file before overwritingin 16 of 674, across 8 files
keep the memory index under 200 linesin 15 of 674
Provide actionable steps and verificationin 13 of 674, across 2 files

Said here and by no other author read

estimate tokens per always-loaded component
classify each component into always needed, sometimes, or rarely
flag oversubscribed mcp servers
flag bloated agent descriptions over thirty words
flag bloated claude.md over one hundred fifty lines
report components ranked by token savings

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.