Agents
AID runs 9 specialized agents across three model tiers. The separation is structural: the reviewer’s tier is always ≥ the executor’s, and the agent that writes code never grades its own work. Each profile maps these tiers to concrete models (see the Agent Model). Generated from canonical/agents/.
Large tier
Section titled “Large tier”Highest-stakes work — requirements, architecture, brownfield discovery, and adversarial review.
aid-architect
Section titled “aid-architect”Tools: Read, Glob, Grep, Write, Edit, Bash
Transforms requirements, SPEC, and KB into design output — SPEC sections, typed dependency-ordered task breakdowns, feature decomposition, delivery sequencing, and DESIGN-typed task execution including UX and flow advice.
Definition: canonical/agents/aid-architect/AGENT.md
Medium tier
Section titled “Medium tier”The production workhorses — implementation, delivery, coordination, and documentation.
aid-developer
Section titled “aid-developer”Tools: Read, Glob, Grep, Write, Edit, Bash
The only agent that modifies production code. Implements TASK files following specs and KB conventions across all implementation task types (IMPLEMENT, TEST, REFACTOR, CONFIGURE, MIGRATE, FIX), with mandatory build verification and formal IMPEDIMENT.md escalation.
Definition: canonical/agents/aid-developer/AGENT.md
aid-interviewer
Section titled “aid-interviewer”Tools: Read, Glob, Grep, Bash
Conducts adaptive one-question-at-a-time dialogue with human stakeholders to gather requirements, clarify ambiguity, and produce REQUIREMENTS.md or Q&A entries.
Definition: canonical/agents/aid-interviewer/AGENT.md
aid-operator
Section titled “aid-operator”Tools: Read, Glob, Grep, Bash, Write
Runs final release verification, packages artifacts, creates PRs and release notes, manages releases, and updates the KB on ship. Safety-first, verification-focused.
Definition: canonical/agents/aid-operator/AGENT.md
aid-orchestrator
Section titled “aid-orchestrator”Tools: Read, Glob, Grep, Bash
Routes pipeline findings to the next phase or skill, enforces human gates, dispatches agents with context, and manages parallel execution.
Definition: canonical/agents/aid-orchestrator/AGENT.md
aid-researcher
Section titled “aid-researcher”Tools: Read, Glob, Grep, Bash, Write, WebSearch, WebFetch
Reads and analyzes code, docs, logs, APIs, and external web sources to produce structured Knowledge Base documents and analysis reports — covering existing-state cataloguing, dependency/integration/convention mapping, telemetry interpretation, security analysis, performance profiling, and web-sourced prior art. Web sources are cited with a URL and access date.
Definition: canonical/agents/aid-researcher/AGENT.md
aid-reviewer
Section titled “aid-reviewer”Tools: Read, Glob, Grep, Bash
Adversarial quality evaluator. Reviews any artifact (code, tasks, specs, plans, KB docs) against its acceptance criteria, rubric, and KB conventions. Produces the 7-column issue ledger with source and severity tags. Does NOT fix anything; does NOT compute the grade.
Definition: canonical/agents/aid-reviewer/AGENT.md
aid-tech-writer
Section titled “aid-tech-writer”Tools: Read, Glob, Grep, Write, Edit, Bash
Authors user-facing documentation — API docs, changelogs, READMEs, release notes, user guides — and reviews existing docs for quality and accuracy.
Definition: canonical/agents/aid-tech-writer/AGENT.md
Small tier
Section titled “Small tier”Deterministic, mechanical operations — extract, format, enumerate.
aid-clerk
Section titled “aid-clerk”Tools: Read, Glob, Grep, Write, Edit, Bash
INTERNAL UTILITY (sub-agent only — do NOT invoke from a skill). Performs one mechanical, schema-bounded operation per dispatch — file extraction, template placeholder-fill, or glob enumeration — returning a markdown table or file with path and line evidence.