Skip to content

Agents

Agents are sub-Claude personas dispatched via the Task tool. Each one ships as agents/<name>.md — frontmatter declares its model, tool grants, and trigger description; the body is the system prompt.

SiftCoder ships fourteen agents. They pair with skills and with the multi-step commands (/agent, /swarm). Most are read-mostly or read-only by design — production-changes happen through coder, qa-fixer, apex-bulkifier, and documenter.

Index

Agent Model Tools Mode
analyst sonnet Read, Grep, Glob, WebFetch, Bash Read-only
apex-bulkifier sonnet Read, Edit, Grep, Glob, Bash Read-write (Apex only)
bridge-analyzer sonnet Read, Grep, Glob, Bash, WebFetch Read-only
coder sonnet Read, Edit, Write, Bash, Grep, Glob Read-write
documenter sonnet Read, Edit, Write, Grep, Glob, Bash Read-write (docs only)
investigator sonnet Read, Grep, Glob, Bash, WebFetch Read-only
lwc-debugger sonnet Read, Grep, Glob, Bash, Edit Read-mostly
memory-curator sonnet Bash, Read Read-write (memory rows only)
orchestrator sonnet Read, Bash, Grep, Glob Read-only
planner sonnet Read, Grep, Glob, Bash, WebFetch Read-only
qa-fixer sonnet Read, Edit, Write, Bash, Grep, Glob Read-write (QA-bounded)
qa-reviewer sonnet Read, Grep, Glob, Bash Read-only
reviewer sonnet Read, Grep, Glob, Bash, WebFetch Read-only
salesforce-architect sonnet Read, Grep, Glob, Bash, WebFetch Read-only
tester sonnet Read, Edit, Write, Grep, Glob, Bash Read-write (tests only)

All ship with model: sonnet — that's the deliberate choice for most reasoning work. If you need cheaper, set model: haiku in your project's overlay; for harder reasoning, lift to opus. There's no per-agent automatic model selection.

analyst

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/analyst.md Tools: Read, Grep, Glob, WebFetch, Bash Model: sonnet

Non-code analyst. Reads specs, PRDs, user research, support tickets, sales transcripts, log dumps, CSV/JSON datasets. Extracts insight; finds patterns; surfaces contradictions. Sibling to investigator, which targets code.

Use it when: the input is not source — a Notion doc, a transcript, a CSV of complaints, a Linear export.

apex-bulkifier

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/apex-bulkifier.md Tools: Read, Edit, Grep, Glob, Bash Model: sonnet

Targeted Apex bulkification specialist. Only job: find SOQL/DML in loops; refactor to bulk-safe collections; preserve behaviour. Returns a diff and a justification. Refuses unrelated changes.

Use it when: a code review flagged "this isn't bulk-safe" and you want a surgical fix.

bridge-analyzer

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/bridge-analyzer.md Tools: Read, Grep, Glob, Bash, WebFetch Model: sonnet

Cross-codebase integration architect. Maps two repos (or services). Generates a gap inventory, picks an integration pattern, drafts a bridge spec. Read-only on both sides.

Use it when: paired with /siftcoder:bridge — two repos that need to talk and you're scoping the contract.

coder

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/coder.md Tools: Read, Edit, Write, Bash, Grep, Glob Model: sonnet

Senior implementer. One job: implement an approved plan, exactly. Surfaces deviations for user approval before deviating; refuses silent scope creep.

Use it when: dispatched by /siftcoder:agent after the planner's output is approved. You can also use it standalone with a hand-written plan.

documenter

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/documenter.md Tools: Read, Edit, Write, Grep, Glob, Bash Model: sonnet

Generates docs in named shapes — architecture, API ref, user manual, technical, runbook, migration. Cites every claim. Pairs with the document skill.

Use it when: you've already decided which shape; you want it filled with grounded evidence.

investigator

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/investigator.md Tools: Read, Grep, Glob, Bash, WebFetch Model: sonnet

Diagnostician. Read-only. Hypothesis-driven. Evidence-cited. Pairs with the investigate skill.

Use it when: a bug, performance issue, or behaviour mystery. Refuses to write code; produces an evidence ledger and ranked hypotheses instead.

lwc-debugger

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/lwc-debugger.md Tools: Read, Grep, Glob, Bash, Edit Model: sonnet

LWC specialist. Isolates causes for wire-adapter no-fires, lifecycle-ordering bugs, reactivity misses, performance regressions. Read-mostly — proposes fixes, doesn't blanket-rewrite.

Use it when: dispatched by /siftcoder:lwc debug <name> or directly when an LWC is misbehaving and the cause isn't obvious.

memory-curator

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/memory-curator.md Tools: Bash, Read Model: sonnet

Librarian for the memory store. Dedupes, merges, prunes. Read-write but bounded — operates only on memory rows, not user code.

Use it when: /siftcoder:mem prune --confirm. Periodically (monthly is plenty) or when search results feel cluttered.

orchestrator

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/orchestrator.md Tools: Read, Bash, Grep, Glob Model: sonnet

Multi-agent orchestrator. Takes a plan, turns it into safe parallel agent dispatch with file-locking, dependency batching, and conflict detection. Spawns workers, merges results.

Use it when: dispatched internally by /siftcoder:agent and /siftcoder:swarm. You don't usually call it directly.

planner

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/planner.md Tools: Read, Grep, Glob, Bash, WebFetch Model: sonnet

Senior planner. Produces plans engineers can execute without re-deciding what's been decided — evidence ledger, risk register, rollback path. Read-only. Beats the native Plan agent for project-context-aware design with memory grounding.

Use it when: any non-trivial change. The first step in /siftcoder:agent.

qa-fixer

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/qa-fixer.md Tools: Read, Edit, Write, Bash, Grep, Glob Model: sonnet

Fix-only implementer. The qa-reviewer report is the scope contract — refuses to expand beyond it. Pairs with qa-reviewer.

Use it when: after qa-reviewer flags issues; you want bounded fixes, not a re-architect.

qa-reviewer

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/qa-reviewer.md Tools: Read, Grep, Glob, Bash Model: sonnet

QA reviewer. Validates implementation against acceptance criteria. Runs tests, checks behaviour vs AC, produces structured pass/fail report. Read-only.

Use it when: the third stage of /siftcoder:agent. Also runnable on demand to AC-check a feature you didn't build with the agent.

reviewer

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/reviewer.md Tools: Read, Grep, Glob, Bash, WebFetch Model: sonnet

Code reviewer with deep project context awareness. Memory-aware (prior decisions), convention-aware. Complementary to native /review — that one covers general best-practice; this one covers project specifics.

Use it when: /siftcoder:review for PR or pending change review.

salesforce-architect

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/salesforce-architect.md Tools: Read, Grep, Glob, Bash, WebFetch Model: sonnet

Principal Salesforce architect. Honest, evidence-based assessments grounded in actual org metadata visible in the workspace. Capacity, sharing, integrations, tech debt. Read-only — written assessment, never modifies metadata.

Use it when: /siftcoder:sf-architect for a periodic org review.

tester

Source: /Users/sam/Documents/Plugins/SiftCoder/agents/tester.md Tools: Read, Edit, Write, Grep, Glob, Bash Model: sonnet

Test specialist. Writes tests that assert behaviour, not implementation. Knows when to mock vs not. Pairs with tdd and fuzz-mind skills.

Use it when: dedicated test-generation pass — usually via /siftcoder:test or as the second half of a TDD loop.

Why all sonnet?

The agents target reasoning quality with predictable cost. Haiku is too rough for plan/review/architect work; opus is overkill for most. The model stays uniform so the orchestrator can dispatch in parallel without tuning per-agent. Override in your project overlay if your cost or quality target differs.