Skip to content

docs: agent prompt design decisions#19

Merged
mpawliszyn merged 1 commit intomainfrom
agent-prompt-design-decisions
Feb 27, 2026
Merged

docs: agent prompt design decisions#19
mpawliszyn merged 1 commit intomainfrom
agent-prompt-design-decisions

Conversation

@mpawliszyn
Copy link
Copy Markdown
Collaborator

Summary

  • Captures design decisions from thorough review of worker agent prompts (Task 7)
  • Covers tool access, output format, version tracking, tone, and behavioral controls
  • Includes 12 design principles derived from experiments and research
  • Q1-Q5 decided, Q6-Q8 pending finalization with further research

Key Decisions

  • Workers stay restricted (no Bash) -- mechanical enforcement via tool restriction
  • Concept-researcher gets Bash + Task for version discovery and routing
  • No per-reference version tracking in V1 -- tree SHA covers primary repo
  • Output templates are the strongest quality lever (evidence-backed)
  • Light anti-rationalization: thoroughness items specific to each agents failure modes

Test plan

  • Review decisions for completeness
  • Confirm Q7 (tone) and Q8 (change boundary) in next session
  • Incorporate behavioral control research findings into Q6

Captures decisions from thorough design review of worker agent prompts
(codebase-locator, codebase-analyzer, codebase-pattern-finder). Includes
tool access decisions, output format guidance, and 12 design principles
derived from experiments and research.

Q1-Q5 decided, Q6-Q8 pending finalization.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@mpawliszyn mpawliszyn merged commit 0391f98 into main Feb 27, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant