Skip to content
Igor Ferreira

Evidence panel

Mesa dos 28

28 guides of good practices for AI-assisted development, each based on the published work of an expert, are turned by code into reviewer subagents for Claude Code, Codex and Antigravity. The table convenes 3 to 5 lenses per topic, and a moderator consolidates a backlog with acceptance criteria. The guarantees live in the code, not in the prompt.

Track
LLMs and agents
Role
Author, personal project
Period
2026
Status
in progress
Tools
Python, Claude Code, Codex, Antigravity, pytest, mypy, ruff

Evidence

  • Citations checked on disk

    Each file:line citation is checked against the cited file; one that does not match gets a correction and, if it is still false, is dropped from the review.

    Measured.

  • 85 test functions, with mutation testing

    pytest with ruff and mypy; mutation testing checks whether the tests catch defects introduced on purpose.

    Measured.

  • Question topic with 85% accuracy (ECE 0.08)

    laya-tema, evaluated a single time on a locked final set of 108 questions; on the same test, the keyword rule got 50% right.

    Measured. See details

  • Effectiveness per round, with real defects, false alarms and cost

    Pre-registered protocol with 10 seeded defects and 3 decoys; the rounds are awaiting approval.

    To be measured.

How it works

Each lens is a subagent generated by code from a guide with a citation, a link and a date. For each topic, the table convenes 3 to 5 lenses, and a moderator merges the reviews into a backlog with acceptance criteria. The final decision stays with me.

The guarantees live in the code:

  • file:line citations checked against the disk, with one correction and then removal;
  • a token budget, with a refusal above the cap;
  • the content of the reviewed project treated as data, against indirect prompt injection;
  • secret masking, with an assertion at the end;
  • mutation testing on the tests themselves.

The guides are updated every Friday by a scheduled task that fetches each expert’s published sources. The topic of each question is suggested by laya-tema.

Credits and notes

  • The guides synthesize the experts' public content; none of them has endorsed the project.