forked from awslabs/amazon-bedrock-agent-samples
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtasks.yaml
More file actions
112 lines (106 loc) · 7.24 KB
/
Copy pathtasks.yaml
File metadata and controls
112 lines (106 loc) · 7.24 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
research_task:
description: >
We are working on the following project: {project_description}.
Conduct research about the target customers and competitors
for the new product or service, while also taking into account the new web domain name
called: {web_domain}. Make sure you perform thorough research to find interesting
and relevant information that will help produce the best possible marketing strategy.
Since you are the first step in the project, you also use the agent store to
save the project description and web domain name.
expected_output: >
A complete report on the target customer base and the competitors for the new
product or service. Provides clarity about the customer demographics, their preferences,
the market positioning versus the competition, and how the target audience gets engaged.
Capture the entire resarch output as json in the agent store with key 'market_research'.
Also the project details are saved as JSON in the agent store with key 'project_details'.
marketing_strategy_task:
description: >
Formulate a comprehensive marketing strategy for this project:
{project_description}, which will have the new web domain name called: {web_domain}.
You must be given as input the entire research results that the Analyst completed
for you (use the agent store for that). Use that research and add your own insights
to create a high-quality strategy.
expected_output: >
A detailed marketing strategy document that outlines the goals, target
audience, key messages, and proposed tactics. Be sure to also identify
marketing channels, and key metrics. The strategy takes into account the
research from your analyst. You save the entire strategy as json
in the agent store with key 'marketing_strategy'.
campaign_idea_task:
description: >
Develop creative marketing campaign ideas for this project: {project_description}.
You must be given as input the entire research results that the Strategist completed.
Ensure your campaign ideas are innovative, engaging, and aligned with the
marketing strategy. Be sure that at least one of the ideas is an
online video ad campaign using a brief but powerful 6-second video.
expected_output: >
A list of three campaign ideas, each with a brief description and
an explanation of the expected impact of the campaign. Sort them by descending order of
positive impact, with the most impactful campaign first. You MUST save the list of
campaign ideas as json in the agent store with key 'campaign_ideas'.
copy_creation_task:
description: >
Create high-level marketing copy for each campaign idea for the project: {project_description}.
You must be given as input the set of campaign ideas created earlier.
Ensure each campaign copy is compelling, clear, and tailored to the target audience.
expected_output: >
For each marketing campaign, output includes: a compelling title, a one line summary,
and importantly adds a few paragraphs of copy. You MUST save the list (in the same order
you received the ideas) of campaign copies as json in the agent store with key 'campaign_copies'.
detailed_campaign_task:
description: >
Pick the first campaign idea from the set of ideas already created. This
is a campaign to help with this project: {project_description}.
For this one campaign idea, draft a detailed Campaign Report. Be as
thorough as possible.
expected_output: >
A detailed campaign report covering only the highest impact campaign idea,
organized with ALL of the following MANDATORY report sections:
A/ CAMPAIGN DESCRIPTION - explain the concept, the audience, the channels, and the assets.
B/ CAMPAIGN VALUE PROPOSITION - explain the product value propisition that is delivered to the audience.
C/ CAMPAIGN CALL TO ACTION - explain what the audience is supposed to do in response to this campaign.
D/ CAMPAIGN METRICS - explain how the effectiveness of the campaign will be measured.
You MUST save the report as json in the agent store with key 'detailed_campaign_report_v1'.
iterative_revisions_task:
description: >
The strategist hands you the draft detailed copy for one campaign as input.
You then go through {feedback_iteration_count} iterations of feedback and revisions to improve the
campaign report and make it more complete.
For each iteration, you ask the Creative Director to provide feedback.
The Director retrieves the latest version of the report AND MUST retrieve the
prior feedback if this is not the first iteration. Based on those inputs
(prior feedback, current report), the Director produces and MUST SAVE a new round of feedback
to the agent store. The Director NEVER forgets to SAVE the feedback to the agent store.
The Writer retrieves the new feedback AND the prior version of the report,
and creates and saves a new version of the report incorporating the feedback.
The Writer NEVER forgets that it must retrieve not only the feedback, but ALSO the prior
version of the report. Without the prior version, the Writer will not be able to
incorporate the feedback into the report and will just start from scratch every time.
Do this until you have produced a final version of the report that incorporates
{feedback_iteration_count} rounds of feedback. To avoid potential confusion, you
give each collaborator the names of the keys to use.
Be sure to enforce all mandatory
report sections. The CEO gets upset if the report is incomplete.
expected_output: >
The final detailed Campaign Report after having incorporated {feedback_iteration_count}
separate rounds of feedback to extend and improve the initial version of the report.
Be sure to mention the key name that was used when saving the final report in the agent store.
final_report_output_task:
description: >
The team of collaborators has done a great job! Now the humans running the actual project
need a polished final report they can use as an excellent starting point to help
our startup customer that is looking to go big with their new product. Give all the agent store
keys for the whole project (including the feedback keys) to the Formatted Report Writer, and have them produce a formatted
final report. The Formatted Report Writer is an expert at doing this. To be more efficient,
break up the work into multiple tasks, and have the writer write the final report in
multiple parts that can be assembled offline. Since the sections are independent, feel free to write them
in parallel. Name these with a 'section_' prefix (e.g., 'section_executive_summary').
As a final step, ask the Formatted Report Writer
to retrieve a fresh copy of the markdown for each section from the agent store.
With all the full text freshly retrieved, concatenate the sections in the correct order,
and save the FULL and complete text (not just placeholders) to the agent store with
key 'complete_final_report'.
expected_output: >
A formatted report saved in the agent store, the keys that its sections are stored in,
and the complete final report key. The complete final report is the concatenation of the
FULL TEXT of all the sections in the correct order.