active_harness 0.2.12 → 0.2.14
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/lib/active_harness/agent/cost.rb +28 -0
- data/lib/active_harness/agent.rb +4 -1
- data/lib/active_harness/costs.rb +14 -3
- data/lib/active_harness/result.rb +2 -1
- data/lib/active_harness/tribunal/dsl.rb +21 -0
- data/lib/active_harness/tribunal/hooks.rb +87 -0
- data/lib/active_harness/tribunal.rb +41 -96
- metadata +4 -1
checksums.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
SHA256:
|
|
3
|
-
metadata.gz:
|
|
4
|
-
data.tar.gz:
|
|
3
|
+
metadata.gz: 6a42e03d329e638e3119d0a259a2286f05a0ec1603b8c2c81636e4ac041ebf9d
|
|
4
|
+
data.tar.gz: 074cf7e1558d9efee20cd17c3405665be00ef29c30574289b54d6e2251ebeacc
|
|
5
5
|
SHA512:
|
|
6
|
-
metadata.gz:
|
|
7
|
-
data.tar.gz:
|
|
6
|
+
metadata.gz: 3397bd1901b3062030b3a41dc327ed1fd7053ef5d90e8e53d0ebb820375054f112cc80e70e528282703277e90137310318d026c456950f86ae755a09aca4e473
|
|
7
|
+
data.tar.gz: 7137703c7469a5f1cd8adfb121fd5047231387890756325f7e035955c7a23dab3bd0dfaf22c223343a9fb006279b6b352c01f1cf443da6cfcf60c3e26227cac3
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
module ActiveHarness
|
|
2
|
+
class Agent
|
|
3
|
+
private
|
|
4
|
+
|
|
5
|
+
# Calculates the monetary cost of a single request based on token usage
|
|
6
|
+
# and pricing data from ActiveHarness::Costs.
|
|
7
|
+
#
|
|
8
|
+
# Returns a hash { input_cost:, output_cost:, total_cost: } in USD,
|
|
9
|
+
# or nil if usage is absent or the model is not found in the pricing registry.
|
|
10
|
+
def calculate_cost(model_id, usage)
|
|
11
|
+
return nil if model_id.nil? || usage.nil?
|
|
12
|
+
|
|
13
|
+
pricing = ActiveHarness::Costs.find(model_id.to_s)
|
|
14
|
+
return nil unless pricing&.input_per_million && pricing&.output_per_million
|
|
15
|
+
|
|
16
|
+
input_cost = (usage[:input_tokens].to_f / 1_000_000) * pricing.input_per_million
|
|
17
|
+
output_cost = (usage[:output_tokens].to_f / 1_000_000) * pricing.output_per_million
|
|
18
|
+
|
|
19
|
+
{
|
|
20
|
+
input_cost: input_cost.round(8),
|
|
21
|
+
output_cost: output_cost.round(8),
|
|
22
|
+
total_cost: (input_cost + output_cost).round(8)
|
|
23
|
+
}
|
|
24
|
+
rescue StandardError
|
|
25
|
+
nil
|
|
26
|
+
end
|
|
27
|
+
end
|
|
28
|
+
end
|
data/lib/active_harness/agent.rb
CHANGED
|
@@ -113,6 +113,7 @@ module ActiveHarness
|
|
|
113
113
|
def build_result(response, entry, attempts, elapsed)
|
|
114
114
|
raw = response[:content]
|
|
115
115
|
parsed = parse_output(raw)
|
|
116
|
+
usage = response[:usage]
|
|
116
117
|
|
|
117
118
|
Result.new(
|
|
118
119
|
input: @input,
|
|
@@ -125,7 +126,8 @@ module ActiveHarness
|
|
|
125
126
|
model_list: model_list,
|
|
126
127
|
attempts: attempts,
|
|
127
128
|
execution_time: elapsed,
|
|
128
|
-
usage:
|
|
129
|
+
usage: usage,
|
|
130
|
+
cost: calculate_cost(entry[:model], usage)
|
|
129
131
|
)
|
|
130
132
|
end
|
|
131
133
|
|
|
@@ -152,4 +154,5 @@ require_relative "agent/models"
|
|
|
152
154
|
require_relative "agent/providers"
|
|
153
155
|
require_relative "agent/output_parser"
|
|
154
156
|
require_relative "agent/ruby_llm_backend"
|
|
157
|
+
require_relative "agent/cost"
|
|
155
158
|
|
data/lib/active_harness/costs.rb
CHANGED
|
@@ -183,10 +183,21 @@ module ActiveHarness
|
|
|
183
183
|
end
|
|
184
184
|
|
|
185
185
|
def registry
|
|
186
|
-
@registry ||=
|
|
187
|
-
|
|
188
|
-
|
|
186
|
+
@registry ||= load_registry
|
|
187
|
+
end
|
|
188
|
+
|
|
189
|
+
def load_registry
|
|
190
|
+
if File.exist?(cache_file)
|
|
191
|
+
begin
|
|
192
|
+
data = JSON.parse(File.read(cache_file), symbolize_names: true)
|
|
193
|
+
return data if data.is_a?(Array)
|
|
194
|
+
rescue JSON::ParserError
|
|
195
|
+
# Cache file corrupted — fall through to bundled data
|
|
196
|
+
end
|
|
189
197
|
end
|
|
198
|
+
JSON.parse(File.read(BUNDLED_DATA_FILE), symbolize_names: true)
|
|
199
|
+
rescue JSON::ParserError, Errno::ENOENT
|
|
200
|
+
[]
|
|
190
201
|
end
|
|
191
202
|
|
|
192
203
|
def fetch_models_dev
|
|
@@ -6,5 +6,6 @@ module ActiveHarness
|
|
|
6
6
|
# output — raw string from the provider
|
|
7
7
|
# parsed — for format :json: a Ruby Hash/Array; for format :text: same as output
|
|
8
8
|
# usage — token counts: { input_tokens:, output_tokens:, total_tokens: } or nil for streaming
|
|
9
|
-
|
|
9
|
+
# cost — { input_cost:, output_cost:, total_cost: } in USD, or nil if pricing unavailable
|
|
10
|
+
Result = Struct.new(:input, :output, :parsed, :system_prompt, :provider, :model, :temperature, :model_list, :attempts, :execution_time, :usage, :cost, keyword_init: true)
|
|
10
11
|
end
|
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
module ActiveHarness
|
|
2
|
+
class Tribunal
|
|
3
|
+
class << self
|
|
4
|
+
# Declare agents at the class level.
|
|
5
|
+
#
|
|
6
|
+
# agents PolitenessAgent, ConstructivenessAgent
|
|
7
|
+
# agents [PolitenessAgent, ConstructivenessAgent]
|
|
8
|
+
def agents(*list)
|
|
9
|
+
tribunal_config[:agents] = list.flatten
|
|
10
|
+
end
|
|
11
|
+
|
|
12
|
+
# Class-level process block — defines how to compute the verdict from results.
|
|
13
|
+
#
|
|
14
|
+
# process { |results| results.all? { |r| r.parsed["result"] == true } }
|
|
15
|
+
# process { |results| results.count { |r| r.parsed["result"] == true } >= 2 }
|
|
16
|
+
def process(&block)
|
|
17
|
+
tribunal_config[:process] = block
|
|
18
|
+
end
|
|
19
|
+
end
|
|
20
|
+
end
|
|
21
|
+
end
|
|
@@ -0,0 +1,87 @@
|
|
|
1
|
+
module ActiveHarness
|
|
2
|
+
class Tribunal
|
|
3
|
+
VALID_HOOKS = %i[
|
|
4
|
+
before_call
|
|
5
|
+
before_agent
|
|
6
|
+
after_agent
|
|
7
|
+
agent_error
|
|
8
|
+
after_call
|
|
9
|
+
before_verdict
|
|
10
|
+
after_verdict
|
|
11
|
+
].freeze
|
|
12
|
+
|
|
13
|
+
class << self
|
|
14
|
+
# Class-level hook registration.
|
|
15
|
+
#
|
|
16
|
+
# on :before_call do ... end
|
|
17
|
+
# on :before_agent do |agent| ... end
|
|
18
|
+
# on :after_agent do |result| ... end
|
|
19
|
+
# on :agent_error do |name, error| ... end
|
|
20
|
+
# on :after_call do |results, errors| ... end
|
|
21
|
+
# on :before_verdict do |results| results end # transform hook
|
|
22
|
+
# on :after_verdict do |verdict| ... end
|
|
23
|
+
def on(event, &block)
|
|
24
|
+
unless VALID_HOOKS.include?(event)
|
|
25
|
+
raise ArgumentError,
|
|
26
|
+
"Unknown Tribunal hook :#{event}. Valid hooks: #{VALID_HOOKS.map { |h| ":#{h}" }.join(", ")}"
|
|
27
|
+
end
|
|
28
|
+
|
|
29
|
+
tribunal_config[:hooks][event] = block
|
|
30
|
+
end
|
|
31
|
+
|
|
32
|
+
# Rails-style aliases for +on+:
|
|
33
|
+
#
|
|
34
|
+
# before :call do ... end # → on :before_call
|
|
35
|
+
# before :agent do |agent| end # → on :before_agent
|
|
36
|
+
# before :verdict do |results| end # → on :before_verdict (transform)
|
|
37
|
+
# after :call do |r, e| end # → on :after_call
|
|
38
|
+
# after :agent do |result| end # → on :after_agent
|
|
39
|
+
# after :verdict do |verdict| end # → on :after_verdict
|
|
40
|
+
# callback :agent_error do |name, e| end # → on :agent_error
|
|
41
|
+
def before(event, &block)
|
|
42
|
+
on(:"before_#{event}", &block)
|
|
43
|
+
end
|
|
44
|
+
|
|
45
|
+
def after(event, &block)
|
|
46
|
+
on(:"after_#{event}", &block)
|
|
47
|
+
end
|
|
48
|
+
|
|
49
|
+
def callback(event, &block)
|
|
50
|
+
on(event, &block)
|
|
51
|
+
end
|
|
52
|
+
end
|
|
53
|
+
|
|
54
|
+
# Instance-level hook registration — overrides class-level hooks for this instance.
|
|
55
|
+
# :before_verdict is a transform hook: its return value replaces the results array
|
|
56
|
+
# passed to the process block.
|
|
57
|
+
def on(event, &block)
|
|
58
|
+
unless VALID_HOOKS.include?(event)
|
|
59
|
+
raise ArgumentError,
|
|
60
|
+
"Unknown Tribunal hook :#{event}. Valid hooks: #{VALID_HOOKS.map { |h| ":#{h}" }.join(", ")}"
|
|
61
|
+
end
|
|
62
|
+
|
|
63
|
+
@hooks[event] = block
|
|
64
|
+
self
|
|
65
|
+
end
|
|
66
|
+
|
|
67
|
+
private
|
|
68
|
+
|
|
69
|
+
def run_hook(event, *args)
|
|
70
|
+
return unless @hooks[event]
|
|
71
|
+
|
|
72
|
+
if args.any?
|
|
73
|
+
instance_exec(*args, &@hooks[event])
|
|
74
|
+
else
|
|
75
|
+
instance_eval(&@hooks[event])
|
|
76
|
+
end
|
|
77
|
+
end
|
|
78
|
+
|
|
79
|
+
# Like run_hook but uses the return value to replace the passed value.
|
|
80
|
+
# Used by :before_verdict to allow results transformation before verdict computation.
|
|
81
|
+
def transform_hook(event, value)
|
|
82
|
+
return value unless @hooks[event]
|
|
83
|
+
|
|
84
|
+
instance_exec(value, &@hooks[event])
|
|
85
|
+
end
|
|
86
|
+
end
|
|
87
|
+
end
|
|
@@ -29,62 +29,11 @@ module ActiveHarness
|
|
|
29
29
|
# ContentQualityTribunal.new(input: "...").call
|
|
30
30
|
#
|
|
31
31
|
class Tribunal
|
|
32
|
-
VALID_HOOKS = %i[
|
|
33
|
-
before_call
|
|
34
|
-
after_agent
|
|
35
|
-
agent_error
|
|
36
|
-
after_call
|
|
37
|
-
before_verdict
|
|
38
|
-
after_verdict
|
|
39
|
-
].freeze
|
|
40
|
-
|
|
41
32
|
# -------------------------------------------------------------------------
|
|
42
|
-
# Class-level DSL —
|
|
33
|
+
# Class-level DSL — core
|
|
43
34
|
# -------------------------------------------------------------------------
|
|
44
35
|
class << self
|
|
45
|
-
#
|
|
46
|
-
# agents PolitenessAgent, ConstructivenessAgent
|
|
47
|
-
def agents(*list)
|
|
48
|
-
tribunal_config[:agents] = list.flatten
|
|
49
|
-
end
|
|
50
|
-
|
|
51
|
-
# Class-level hook registration.
|
|
52
|
-
# on(:after_agent) { |result| puts result.model }
|
|
53
|
-
def on(event, &block)
|
|
54
|
-
unless VALID_HOOKS.include?(event)
|
|
55
|
-
raise ArgumentError, "Unknown Tribunal hook :#{event}. Valid hooks: #{VALID_HOOKS.join(", ")}"
|
|
56
|
-
end
|
|
57
|
-
|
|
58
|
-
tribunal_config[:hooks][event] = block
|
|
59
|
-
end
|
|
60
|
-
|
|
61
|
-
# Rails-style aliases for +on+:
|
|
62
|
-
#
|
|
63
|
-
# before :call do ... end # → on :before_call
|
|
64
|
-
# before :agent do ... end # → on :before_agent (not used yet)
|
|
65
|
-
# before :verdict do |r| end # → on :before_verdict
|
|
66
|
-
# after :call do ... end # → on :after_call
|
|
67
|
-
# after :agent do |r| end # → on :after_agent
|
|
68
|
-
# after :verdict do |v| end # → on :after_verdict
|
|
69
|
-
# callback :agent_error do |n,e| end # → on :agent_error
|
|
70
|
-
def before(event, &block)
|
|
71
|
-
on(:"before_#{event}", &block)
|
|
72
|
-
end
|
|
73
|
-
|
|
74
|
-
def after(event, &block)
|
|
75
|
-
on(:"after_#{event}", &block)
|
|
76
|
-
end
|
|
77
|
-
|
|
78
|
-
def callback(event, &block)
|
|
79
|
-
on(event, &block)
|
|
80
|
-
end
|
|
81
|
-
|
|
82
|
-
# Class-level process block.
|
|
83
|
-
# process { |results| results.all? { |r| r.parsed["result"] == true } }
|
|
84
|
-
def process(&block)
|
|
85
|
-
tribunal_config[:process] = block
|
|
86
|
-
end
|
|
87
|
-
|
|
36
|
+
# Each subclass gets its own isolated config hash.
|
|
88
37
|
def tribunal_config
|
|
89
38
|
@tribunal_config ||= { agents: [], hooks: {} }
|
|
90
39
|
end
|
|
@@ -94,36 +43,32 @@ module ActiveHarness
|
|
|
94
43
|
end
|
|
95
44
|
end
|
|
96
45
|
|
|
97
|
-
|
|
46
|
+
# -------------------------------------------------------------------------
|
|
47
|
+
# Instance API
|
|
48
|
+
# -------------------------------------------------------------------------
|
|
49
|
+
attr_accessor :input, :context, :stream, :agent_event_stream, :tribunal_event_stream
|
|
98
50
|
attr_reader :results, :errors, :verdict, :execution_time, :agent_execution_times
|
|
99
51
|
|
|
100
|
-
def initialize(input: nil, context: {}, agents: nil, timeout: 7
|
|
52
|
+
def initialize(input: nil, context: {}, agents: nil, timeout: 7,
|
|
53
|
+
stream: nil, agent_event_stream: nil, tribunal_event_stream: nil)
|
|
101
54
|
config = self.class.tribunal_config
|
|
102
55
|
|
|
103
|
-
@input
|
|
104
|
-
@context
|
|
105
|
-
@agents
|
|
106
|
-
@timeout
|
|
107
|
-
@process_block
|
|
108
|
-
@hooks
|
|
109
|
-
@
|
|
110
|
-
@
|
|
111
|
-
@
|
|
112
|
-
@
|
|
56
|
+
@input = input
|
|
57
|
+
@context = context
|
|
58
|
+
@agents = agents || config[:agents]
|
|
59
|
+
@timeout = timeout
|
|
60
|
+
@process_block = config[:process]
|
|
61
|
+
@hooks = config[:hooks].dup
|
|
62
|
+
@stream = stream
|
|
63
|
+
@agent_event_stream = agent_event_stream
|
|
64
|
+
@tribunal_event_stream = tribunal_event_stream
|
|
65
|
+
@results = []
|
|
66
|
+
@errors = []
|
|
67
|
+
@verdict = nil
|
|
68
|
+
@execution_time = nil
|
|
113
69
|
@agent_execution_times = []
|
|
114
70
|
end
|
|
115
71
|
|
|
116
|
-
# Instance-level hook registration — overrides class-level hooks.
|
|
117
|
-
# :before_verdict is a transform hook: its return value replaces the results array.
|
|
118
|
-
def on(event, &block)
|
|
119
|
-
unless VALID_HOOKS.include?(event)
|
|
120
|
-
raise ArgumentError, "Unknown Tribunal hook :#{event}. Valid hooks: #{VALID_HOOKS.join(", ")}"
|
|
121
|
-
end
|
|
122
|
-
|
|
123
|
-
@hooks[event] = block
|
|
124
|
-
self
|
|
125
|
-
end
|
|
126
|
-
|
|
127
72
|
# Instance-level process block — overrides class-level block.
|
|
128
73
|
def process(&block)
|
|
129
74
|
@process_block = block
|
|
@@ -143,14 +88,15 @@ module ActiveHarness
|
|
|
143
88
|
|
|
144
89
|
started_at = Process.clock_gettime(Process::CLOCK_MONOTONIC)
|
|
145
90
|
|
|
146
|
-
futures = agents.map do |agent|
|
|
91
|
+
futures = agents.each_with_index.map do |agent, index|
|
|
92
|
+
run_hook(:before_agent, agent, index)
|
|
147
93
|
t0 = Process.clock_gettime(Process::CLOCK_MONOTONIC)
|
|
148
94
|
future = Concurrent::Future.execute { agent.call }
|
|
149
95
|
[future, t0]
|
|
150
96
|
end
|
|
151
97
|
|
|
152
|
-
@results
|
|
153
|
-
@errors
|
|
98
|
+
@results = []
|
|
99
|
+
@errors = []
|
|
154
100
|
@agent_execution_times = []
|
|
155
101
|
|
|
156
102
|
futures.each_with_index do |(future, t0), index|
|
|
@@ -159,19 +105,19 @@ module ActiveHarness
|
|
|
159
105
|
@agent_execution_times << { agent: agents[index].class.name, time: elapsed }
|
|
160
106
|
|
|
161
107
|
if future.fulfilled?
|
|
162
|
-
value
|
|
108
|
+
value = future.value
|
|
163
109
|
result = value.is_a?(ActiveHarness::Agent) ? value.result : value
|
|
164
110
|
@results << result
|
|
165
|
-
run_hook(:after_agent, result)
|
|
111
|
+
run_hook(:after_agent, result, index)
|
|
166
112
|
elsif future.incomplete?
|
|
167
113
|
error = Errors::TimeoutError.new(
|
|
168
114
|
"Agent #{agents[index].class.name} timed out after #{@timeout}s"
|
|
169
115
|
)
|
|
170
116
|
@errors << { agent: agents[index].class.name, error: error }
|
|
171
|
-
run_hook(:agent_error, agents[index].class.name, error)
|
|
117
|
+
run_hook(:agent_error, agents[index].class.name, error, index)
|
|
172
118
|
else
|
|
173
119
|
@errors << { agent: agents[index].class.name, error: future.reason }
|
|
174
|
-
run_hook(:agent_error, agents[index].class.name, future.reason)
|
|
120
|
+
run_hook(:agent_error, agents[index].class.name, future.reason, index)
|
|
175
121
|
end
|
|
176
122
|
end
|
|
177
123
|
|
|
@@ -193,26 +139,25 @@ module ActiveHarness
|
|
|
193
139
|
|
|
194
140
|
private
|
|
195
141
|
|
|
196
|
-
def run_hook(event, *args)
|
|
197
|
-
@hooks[event]&.call(*args)
|
|
198
|
-
end
|
|
199
|
-
|
|
200
|
-
# Like run_hook but uses the return value to replace the passed value.
|
|
201
|
-
def transform_hook(event, value)
|
|
202
|
-
return value unless @hooks[event]
|
|
203
|
-
|
|
204
|
-
@hooks[event].call(value)
|
|
205
|
-
end
|
|
206
|
-
|
|
207
142
|
def resolve_agents
|
|
208
143
|
@agents.map do |agent|
|
|
209
144
|
if agent.is_a?(Class)
|
|
210
|
-
agent.new(
|
|
145
|
+
agent.new(
|
|
146
|
+
input: @input,
|
|
147
|
+
context: @context.dup,
|
|
148
|
+
stream: @stream,
|
|
149
|
+
event_stream: @agent_event_stream
|
|
150
|
+
)
|
|
211
151
|
else
|
|
212
|
-
agent.input
|
|
152
|
+
agent.input = @input if @input
|
|
153
|
+
agent.stream = @stream if @stream
|
|
154
|
+
agent.event_stream = @agent_event_stream if @agent_event_stream
|
|
213
155
|
agent
|
|
214
156
|
end
|
|
215
157
|
end
|
|
216
158
|
end
|
|
217
159
|
end
|
|
218
160
|
end
|
|
161
|
+
|
|
162
|
+
require_relative "tribunal/hooks"
|
|
163
|
+
require_relative "tribunal/dsl"
|
metadata
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
|
2
2
|
name: active_harness
|
|
3
3
|
version: !ruby/object:Gem::Version
|
|
4
|
-
version: 0.2.
|
|
4
|
+
version: 0.2.14
|
|
5
5
|
platform: ruby
|
|
6
6
|
authors:
|
|
7
7
|
- the-teacher
|
|
@@ -33,6 +33,7 @@ extra_rdoc_files: []
|
|
|
33
33
|
files:
|
|
34
34
|
- lib/active_harness.rb
|
|
35
35
|
- lib/active_harness/agent.rb
|
|
36
|
+
- lib/active_harness/agent/cost.rb
|
|
36
37
|
- lib/active_harness/agent/hooks.rb
|
|
37
38
|
- lib/active_harness/agent/models.rb
|
|
38
39
|
- lib/active_harness/agent/output_parser.rb
|
|
@@ -71,6 +72,8 @@ files:
|
|
|
71
72
|
- lib/active_harness/railtie.rb
|
|
72
73
|
- lib/active_harness/result.rb
|
|
73
74
|
- lib/active_harness/tribunal.rb
|
|
75
|
+
- lib/active_harness/tribunal/dsl.rb
|
|
76
|
+
- lib/active_harness/tribunal/hooks.rb
|
|
74
77
|
- lib/generators/active_harness/agent/agent_generator.rb
|
|
75
78
|
- lib/generators/active_harness/agent/templates/agent.rb.tt
|
|
76
79
|
- lib/generators/active_harness/install/install_generator.rb
|