token-budget-advisor

Name: token-budget-advisor
Author: Svenja-dev

4installs5forks

How to install token-budget-advisor?

svenja-dev/claude-code-skills/token-budget-advisor

$npx -y skills add Svenja-dev/claude-code-skills --skill token-budget-advisor

Installs into the current project.

Summary

Proactive token budget assessment and task chunking strategy. Use this skill when queries involve multiple large file uploads, requests for comprehensive multi-document analysis, complex multi-step workflows with heavy research (10+ tool calls), phrases like "complete analysis",

Files · 1

View on GitHub

SKILL.md

1	# Token Budget Advisor
2
3	This skill provides early assessment of token-heavy tasks and recommends chunking strategies to ensure successful completion within context window constraints.
4
5	## When to Use This Skill
6
7	Trigger this skill before beginning work when you detect:
8
9	- Multiple file uploads (3+ documents) combined with analysis requests
10	- Requests for "comprehensive", "complete", "thorough", or "full" analysis
11	- Multi-document comparative analysis
12	- Complex workflows requiring 10+ tool calls (extensive web research + synthesis)
13	- Tasks combining heavy research with large artifacts (reports, presentations)
14	- Queries spanning multiple dimensions (temporal + categorical + quantitative)
15	- Requests to "analyze everything" or "create a complete report on all aspects"
16
17	## Core Function
18
19	This skill serves two purposes:
20
21	1. Early warning system: Assess whether a task will likely exceed token limits
22	2. Strategic planning: Provide specific, actionable chunking recommendations
23
24	## Token Estimation Framework
25
26	### Quick Assessment Heuristics
27
28	Estimate token consumption using these rough guidelines:
29
30	Input costs:
31	- Uploaded document: ~1,000-5,000 tokens each (depending on length)
32	- Web search result: ~500-1,500 tokens
33	- Web fetch (full article): ~2,000-8,000 tokens
34	- Google Drive document: ~1,000-10,000 tokens (varies significantly)
35
36	Output costs:
37	- Simple response: 500-2,000 tokens
38	- Detailed analysis: 2,000-5,000 tokens
39	- Long-form report: 5,000-15,000 tokens
40	- Complex artifact (presentation, document): 5,000-20,000 tokens
41
42	Tool call overhead:
43	- Each tool call includes the query, results, and reasoning: ~1,000-3,000 tokens average
44
45	Warning thresholds:
46
47	- Caution zone (60-80% of budget): Task is achievable but tight; consider efficiency
48	- Danger zone (80-95% of budget): High risk; strongly recommend chunking
49	- Exceeds budget (95%+ of budget): Task requires chunking; cannot complete in one conversation
50
51	### Task Complexity Multipliers
52
53	Apply these mental adjustments:
54
55	- Synthesis required: Add 30-50% to output estimate (comparing, integrating multiple sources)
56	- Iterative refinement: Add 20-30% (when task involves reviewing and improving)
57	- Multiple formats: Add 20% per additional output type (report + presentation + spreadsheet)
58
59	## Chunking Strategy Framework
60
61	When a task exceeds token budget, recommend specific chunking approaches. Choose strategies based on task structure:
62
63	### 1. Sequential Processing
64
65	Best for: Time-series data, chronological analysis, ordered workflows
66
67	Pattern:
68	```
69	"This analysis of 12 months of data will exceed our token budget. I recommend we split it into quarters:
70	- Part 1: Q1-Q2 analysis (Jan-Jun)
71	- Part 2: Q3-Q4 analysis (Jul-Dec)
72	- Part 3: Synthesis and recommendations
73
74	Should I start with Part 1?"
75	```
76
77	When to use:
78	- Historical data analysis
79	- Period-over-period comparisons
80	- Multi-phase projects
81
82	### 2. Dimensional Breakdown
83
84	Best for: Multi-faceted analysis, different aspects of same topic
85
86	Pattern:
87	```
88	"A complete market analysis covering financial, competitive, regulatory, and technological factors will strain our token budget. Let's break it into:
89	- Session 1: Financial performance and market size
90	- Session 2: Competitive landscape and positioning
91	- Session 3: Regulatory environment and compliance
92	- Session 4: Technology trends and synthesis
93
94	Which dimension should we tackle first?"
95	```
96
97	When to use:
98	- Multi-stakeholder analysis
99	- Different analytical lenses on same subject
100	- Complex business cases
101
102	### 3. Depth Progression
103
104	Best for: Tasks requiring outline → draft → refinement
105
106	Pattern:
107	```
108	"Creating a comprehensive 50-slide presentation with detailed research will exceed our budget. I recommend:
109	- Round 1: Build structure and outline (30 min)
110	- Round 2: Develop content for slides 1-25 (45 min)
111	- Round 3: Develop content for slides 26-50 (45 min)
112	- Round 4: Refinement pass (30 min)
113
114	Let's start with the outline?"
115	```
116
117	When to use:
118	- Large documents or presentations
119	- When quality refinement is important
120	- Creative projects benefiting from iteration
121
122	### 4. Subset Sampling
123
124	Best for: Large document sets where representative sampling works
125
126	Pattern:
127	```
128	"Analyzing all 15 contracts will exceed our budget. I suggest:
129	- Part 1: Analyze 5 representative contracts (different types/dates)
130	- Part 2: Based on patterns found, confirm with 5 more
131	- Part

People who liked this also viewed

6 picks

Type

token-budget-advisor

4installs5forks

View on GitHub

How to install token-budget-advisor?

svenja-dev/claude-code-skills/token-budget-advisor

$npx -y skills add Svenja-dev/claude-code-skills --skill token-budget-advisor

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Token Budget Advisor
2
3	This skill provides early assessment of token-heavy tasks and recommends chunking strategies to ensure successful completion within context window constraints.
4
5	## When to Use This Skill
6
7	Trigger this skill before beginning work when you detect:
8
9	- Multiple file uploads (3+ documents) combined with analysis requests
10	- Requests for "comprehensive", "complete", "thorough", or "full" analysis
11	- Multi-document comparative analysis
12	- Complex workflows requiring 10+ tool calls (extensive web research + synthesis)
13	- Tasks combining heavy research with large artifacts (reports, presentations)
14	- Queries spanning multiple dimensions (temporal + categorical + quantitative)
15	- Requests to "analyze everything" or "create a complete report on all aspects"
16
17	## Core Function
18
19	This skill serves two purposes:
20
21	1. Early warning system: Assess whether a task will likely exceed token limits
22	2. Strategic planning: Provide specific, actionable chunking recommendations
23
24	## Token Estimation Framework
25
26	### Quick Assessment Heuristics
27
28	Estimate token consumption using these rough guidelines:
29
30	Input costs:
31	- Uploaded document: ~1,000-5,000 tokens each (depending on length)
32	- Web search result: ~500-1,500 tokens
33	- Web fetch (full article): ~2,000-8,000 tokens
34	- Google Drive document: ~1,000-10,000 tokens (varies significantly)
35
36	Output costs:
37	- Simple response: 500-2,000 tokens
38	- Detailed analysis: 2,000-5,000 tokens
39	- Long-form report: 5,000-15,000 tokens
40	- Complex artifact (presentation, document): 5,000-20,000 tokens
41
42	Tool call overhead:
43	- Each tool call includes the query, results, and reasoning: ~1,000-3,000 tokens average
44
45	Warning thresholds:
46
47	- Caution zone (60-80% of budget): Task is achievable but tight; consider efficiency
48	- Danger zone (80-95% of budget): High risk; strongly recommend chunking
49	- Exceeds budget (95%+ of budget): Task requires chunking; cannot complete in one conversation
50
51	### Task Complexity Multipliers
52
53	Apply these mental adjustments:
54
55	- Synthesis required: Add 30-50% to output estimate (comparing, integrating multiple sources)
56	- Iterative refinement: Add 20-30% (when task involves reviewing and improving)
57	- Multiple formats: Add 20% per additional output type (report + presentation + spreadsheet)
58
59	## Chunking Strategy Framework
60
61	When a task exceeds token budget, recommend specific chunking approaches. Choose strategies based on task structure:
62
63	### 1. Sequential Processing
64
65	Best for: Time-series data, chronological analysis, ordered workflows
66
67	Pattern:
68	```
69	"This analysis of 12 months of data will exceed our token budget. I recommend we split it into quarters:
70	- Part 1: Q1-Q2 analysis (Jan-Jun)
71	- Part 2: Q3-Q4 analysis (Jul-Dec)
72	- Part 3: Synthesis and recommendations
73
74	Should I start with Part 1?"
75	```
76
77	When to use:
78	- Historical data analysis
79	- Period-over-period comparisons
80	- Multi-phase projects
81
82	### 2. Dimensional Breakdown
83
84	Best for: Multi-faceted analysis, different aspects of same topic
85
86	Pattern:
87	```
88	"A complete market analysis covering financial, competitive, regulatory, and technological factors will strain our token budget. Let's break it into:
89	- Session 1: Financial performance and market size
90	- Session 2: Competitive landscape and positioning
91	- Session 3: Regulatory environment and compliance
92	- Session 4: Technology trends and synthesis
93
94	Which dimension should we tackle first?"
95	```
96
97	When to use:
98	- Multi-stakeholder analysis
99	- Different analytical lenses on same subject
100	- Complex business cases
101
102	### 3. Depth Progression
103
104	Best for: Tasks requiring outline → draft → refinement
105
106	Pattern:
107	```
108	"Creating a comprehensive 50-slide presentation with detailed research will exceed our budget. I recommend:
109	- Round 1: Build structure and outline (30 min)
110	- Round 2: Develop content for slides 1-25 (45 min)
111	- Round 3: Develop content for slides 26-50 (45 min)
112	- Round 4: Refinement pass (30 min)
113
114	Let's start with the outline?"
115	```
116
117	When to use:
118	- Large documents or presentations
119	- When quality refinement is important
120	- Creative projects benefiting from iteration
121
122	### 4. Subset Sampling
123
124	Best for: Large document sets where representative sampling works
125
126	Pattern:
127	```
128	"Analyzing all 15 contracts will exceed our budget. I suggest:
129	- Part 1: Analyze 5 representative contracts (different types/dates)
130	- Part 2: Based on patterns found, confirm with 5 more
131	- Part

People who liked this also viewed

6 picks

Type