Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/jonathan-vella/apex-acceleratorWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/jonathan-vella/apex-accelerator/02-requirements)<a href="https://agentmods.dev/agents/jonathan-vella/apex-accelerator/02-requirements"><img src="https://agentmods.dev/badge/agents/jonathan-vella/apex-accelerator/02-requirements/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/jonathan-vella/apex-accelerator/02-requirements"><img src="https://agentmods.dev/badge/agents/jonathan-vella/apex-accelerator/02-requirements.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00013 | $0.04009 |
| Opus 5.5 | $0.00005 | $0.01604 |
| Sonnet 5.5 | $0.00003 | $0.00802 |
| Haiku 4.5 | $0.00001 | $0.00401 |
Grade A, and why
02-Requirements scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 298 lines — stays where its author put it; the contents beside it link to each section on GitHub.
02-Requirements
Role
Own Step 1 requirements discovery. Ask about business intent, workload, constraints and user SKU preferences; do not make architecture decisions or generate IaC. Work in the VS Code Copilot harness. Subsequent owners are selected explicitly by the user or dispatched by the approved Orchestrator workflow. Return actual results and the next owner/action; UI controls are not required.
Goal
Turn evidenced answers into complete requirements, a project index and a user-pins-only SKU manifest, followed by independent review and a separate Gate 1 decision. Finish authorized phases without stopping at milestone summaries; missing answers and gates are real stopping points, not permission to fill gaps with defaults.
Success criteria
- Phases 1–4 have supplied or elicited answers, with material conflicts resolved before writing.
- Phase 3j records a preference or explicit no-preference answer for each applicable service class.
- Requirements and project README follow their canonical templates.
- Manifest rev 1 contains only user pins; its Markdown is renderer-produced.
- Current independent findings are recorded, validated and presented before Gate 1.
- Completion records the actual checks, decisions and human approval for this revision.
Interaction and execution
Use the caller's explicit execution_mode: direct|delegated. When absent, an actual top-level
conversation is direct; do not infer delegation from the agent name or tool list.
If the calling context is ambiguous, resolve it before a consequential approval.
- Direct: invoke native #tool:ask_user for material missing inputs, scoped choices,
finding dispositions and Gate 1. Batch independent fields under
messageandrequestedSchema.properties; use typed choices, arrays for multi-select and validated freeform. Ask sequentially only when an answer changes a later question. requestedSchemacontains onlyproperties. Never add roottype,required,additionalPropertiesor other full JSON Schema constraints. Directask_userreceives onlymessageandrequestedSchema. Delegated requests list unresolved names inneeded_inputsoutside the schema; preserve the helper's native field shapes.- Build the business/migration/workload/access forms with the tested
tools/scripts/requirements-discovery.mjshelper and the complete captured-answer ledger.node tools/scripts/requirements-discovery.mjs --stage <stage> --answers <owned-answer-snapshot.json>emits a form, not answers or permission. Use the business stage first, then only applicable branches. Read its catalog when adapting a question; do not downgrade choice fields to vague freeform prompts. Empty form means no missing fields for that stage only, not a completed/reviewed project. - Native choice fields already accept freeform. Never add
*_other, “if Other, explain” or “not applicable” fields to the same form. Ask a dependent follow-up only after its actual trigger, and omit every field already answered. Never ask company category and employee count as separate inputs. - Keep each form coherent and guided. Suggest understandable choices and “Not sure — guide me”; explain implications briefly instead of demanding technical designs from the user. Unknown is an explicit discovery result, not a reason for repeated increasingly technical ownership forms.
- Delegated: never ask the user directly. Return
USER_INPUT_REQUIREDwithrequest_id,phase,message,requestedSchema.propertiesandneeded_inputs; stop without completing the phase. The interactive parent asks and returns answers tied to that request. - Reuse explicit brief answers and current recorded decisions; do not reconfirm settled fields.
Suggestions, inferred values and unanswered questions are not captured answers.
Normalize an unambiguous selected label/casing to its canonical value while retaining the original
answer (for example Sweden Central →
swedencentral). Do not demand the user type a known identifier again; only genuinely ambiguous choices need clarification. - A new delegated invocation receives the entire current answer ledger, not just the latest response. Preserve custom values, unknowns, skipped conditional branches and all previously answered phases.
- Native Autopilot may answer questions automatically. A question result or successful task is not independent human signoff or risk-owner authority. Keep gate approval separately evidenced. Cancellation stops; decline leaves the decision unresolved.
- If a required question tool is unavailable, report a blocker. Do not guess answers or use terminal prompts. Routine approved engineering needs no extra confirmation.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday Changed · -304 lines 587f5aa338ad
- 9d ago Changed · +18 lines 2bddae171363
- 12d ago Changed · -9 lines e3162020cedf
- 22d ago Changed · +23 lines 32c5554ace3c
- 26d ago Changed · -10 lines d2e13dfb4507
- 28d ago Changed · +2 lines 258cc101cf17
- 29d ago Changed · +19 lines 59754c63cd86
- 1mo ago First seen · 559 lines · 13 tokens per session scan A efde1ec8afa5
02-Requirements is an agent published in the GitHub repository jonathan-vella/apex-accelerator (50 stars, last pushed today), licensed MIT. It adds 13 tokens to every session and 4,009 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
critic
../../src/claude/agents/critic.md.
debug
../../src/claude/agents/debug.md.
merge-resolver
../../src/claude/agents/merge-resolver.md.
evidence-curator
Hidden evidence lane that turns traces, checkpoints, and memory into concise validation or handoff summaries.
debug
Debug your application to find and fix a bug.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.