Kartik-Doye/Ravengard-AI-Recruiter / agent/skills/frugal
frugal
VerifiedReduces context and token waste while preserving task-critical evidence, source pointers, and the same quality gate. Use when a software task is context-heavy, repetitive, expensive, or likely to exceed an agent context window.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 10.6 KB |
| references/budget-protocol.md | markdown | 3.0 KB |
| references/measurement.md | markdown | 2.1 KB |
| references/research.md | markdown | 826 B |
references/measurement.md
View on GitHub# Measurement rules: the honesty framework for savings claims
Research context (PRD §6.25): token consumption varies enormously
between agents and more tokens do not automatically mean better
outcomes: which is exactly why the KPI is **successful task token
cost**, not raw minimization.
## The four categories
| Category | Meaning | Allowed phrasing |
| --- | --- | --- |
| **Measured** | Runtime/provider-supplied usage numbers compared across runs | "input tokens: 48,200 → 21,700 (usage fields, same task, same model)" |
| **Derived** | Reduction computed from known content with an explicit tokenizer or byte/char proxy | "context bytes read: 812KB → 96KB (−88%, byte counts from file sizes)" |
| **Estimated** | Counterfactual estimate, labeled and based on stated assumptions | "estimated ~40% fewer tokens (assumes median file 300 lines, based on 12 sampled reads)" |
| **Unknown** | Cannot honestly be measured | "savings unknown: no usage telemetry this session" |
**Never** print "saved 63% of tokens" without the category and its
evidence. An uncategorized percentage is a violation of this skill, not
a rounding choice.
## Metrics worth tracking (when the runtime exposes them)
- input tokens / output tokens / cache-read / cache-write;
- context bytes read (files opened, lines retrieved);
- command-output bytes before/after capping;
- tool calls per task; retries per task;
- **task success**: the multiplier on everything: a saving that costs
a retry is a loss.
## Baseline discipline
- Compare like with like: same task, same model, same data; state what
differed.
- One-variable changes (cap the logs OR range-read, not both) when the
goal is attribution; bundle when the goal is just "do the task
cheaper": say which goal.
- Cache-prefix savings only count when the runtime actually caches,
verify, don't assume.
## Where this skill's own budget applies
- Session handoffs and context compaction: load the smallest set of current,
decision-relevant evidence.
- This suite's own SKILL.md files follow the Agent Skills 500-line structural
limit, with conditional detail loaded from references.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |