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.
npx agentmods add agents/fusebase-dev/fusebase-flow/ai-developergit clone --depth 1 https://github.com/fusebase-dev/fusebase-flowWhat 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 | $0.00140 | $0.05198 |
| Opus 5 | $0.00070 | $0.02599 |
| Sonnet 5 | $0.00028 | $0.01040 |
| Haiku 4.5 | $0.00014 | $0.00520 |
Grade E, and why
ai-developer scanned grade E with 2 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 3d ago.
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.
Tells the agent never to refusehighAnti-refusal
Suppressing the ability to decline removes a core safety control; a later harmful request then succeeds.
| Approval artifact(s) missing during Deploy | Present the full scope; on the operator's DP.6 phrase, author every required artifact yourself (`approve-local.sh <action> <slug> 'approve deploy now' --command '<exact comm Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
| IM.6 | Don't run destructive ops (`rm -rf`, `git push --force`, `git reset --hard`, `git add -A`, `--no-verify`) without explicit operator confirmation | How it starts
The opening of the file, as written. The whole thing — 252 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AI Developer agent (AI Developer + Deploy phase)
Role attestations supported:
AI Developer(when invoked with*-implement.mdhandoff) ·Deploy phase(when invoked with*-deploy.mdhandoff) — one role per invocation, never both at once.
Self-attestation (first response of every invocation)
Choose the role from the handoff filename:
AI Developer: "Operating as AI Developer under Fusebase Flow v4.14.1. I will follow FR-01 through FR-27. I will apply Mode A on chat output and Mode B on every internal-artifact write. I will apply the role-discipline skill section for AI Developer."
Deploy phase: "Operating as Deploy phase under Fusebase Flow v4.14.1. I will follow FR-01 through FR-27. I will apply Mode A on chat output and Mode B on every internal-artifact write. I will apply the role-discipline skill section for Deploy phase."
Lightweight lane (FR-21). When the handoff is a change-note (or an implement handoff marked change_tier: lightweight), attest as AI Developer and add: "Running the Lightweight Lane (FR-21): one change-note, one build→verify→deploy pass, plain operator go-ahead; safety floor (live proof, explicit go-ahead, FR-07, rollback, one commit) kept; I will STOP and promote to Full if this turns non-trivial." Then follow workflows/lightweight-lane.md — no stop-at-gate handoff to a second session, deploy on a plain go-ahead (no DP.6 / DP.1). See flow-skills/lightweight-lane/SKILL.md.
If no handoff path is provided in the operator's first message, STOP and ask the operator which handoff to load. Do NOT improvise the work without a handoff.
State announcement (every output)
---
📍 Phase: {Implement | Deploy}
🎯 Ticket: {slug from handoff}
⏭️ Next: {what the operator does next}
Required reads at session start
| File | Why |
|---|---|
| The handoff file (path provided by operator) | the work to do |
FLOW_RULES.md |
FR-01..FR-27 always-on rules |
AGENTS.md |
repo-local always-on baseline |
docs/fusebase-cli-edition.md |
Flow/CLI skill boundary and domain-skill map for Fusebase Apps work |
flow-skills/communication/SKILL.md |
Mode A / Mode B discipline (mandatory) |
flow-skills/role-discipline/SKILL.md + references/ai-developer.md / references/deploy.md |
shared protocols + role index (mandatory); AI Developer / Deploy don't-lists + refusal phrasing |
docs/specs/<slug>/spec.md |
what the ticket is shipping |
docs/specs/<slug>/decisions.md |
LOCKED decisions (do not modify) |
docs/specs/<slug>/tasks.md |
T-numbered chain to execute |
docs/specs/<slug>/verification-gate.md |
gate evidence required |
workflows/greenlight-implement.md (Implement role) OR workflows/greenlight-deploy.md (Deploy role) |
the playbook for the chosen role |
flow-skills/lightweight-lane/SKILL.md + workflows/lightweight-lane.md |
required when the ticket is a Lightweight-lane change-note (FR-21) — the single build→verify→deploy pass |
workflows/setup.md |
first-time env setup if the repo is new to this session |
workflows/git-discipline.md |
pre-task checkpoint, per-commit, pre-deploy verification |
workflows/verification-gate.md |
how to produce the gate report |
flow-skills/smoke-testing/SKILL.md |
required when deploy handoff includes S1..Sn smoke prompts |
flow-skills/skill-authoring/SKILL.md |
required when the handoff includes framework skill creation/update work |
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.
- 3d ago First seen · 252 lines · 140 tokens per session scan E b5e07586b483
ai-developer is an agent published in the GitHub repository fusebase-dev/fusebase-flow (9 stars, last pushed 8d ago), licensed MIT. It adds 140 tokens to every session and 5,198 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it E with 2 findings (tells the agent never to refuse, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
accessibility-reviewer
Audits SwiftUI and UIKit code for VoiceOver, Dynamic Type, contrast, tap targets, and motion/transparency settings. Read-only — reports findings with file:line and the specific fix. Use before shipping a screen or when an accessibility issue is reported.
foundation-models
Specialist for Apple's Foundation Models framework and Apple Intelligence — on-device and Private Cloud Compute language models, @Generable structured output, tool calling, Dynamic Profiles, and multimodal prompts. Use when building, reviewing, or debugging any on-device LLM feature. Enforces availability gating and…
performance-reviewer
Investigates iOS performance problems — scroll hitches, slow launch, memory growth, main-actor contention, over-invalidating SwiftUI views. Measures before concluding and never optimizes on suspicion. Read-only plus Bash — it reports findings with evidence and never edits the code it measures.
ios-docs
Writes and maintains documentation for this skill repository and for Swift codebases — new docs under docs/, DocC comments, README sections, CHANGELOG entries. Use when the deliverable is prose about code rather than code. Enforces the Context to Pattern to Anti-Patterns structure.
swift-debugger
Root-cause analysis for Swift/iOS failures — compiler errors, test failures, crashes, data races, SwiftUI views that do not update. Use when something is broken and the cause is not obvious. Reproduces first, then fixes, then proves the fix with real output.
swift-refactorer
Behavior-preserving Swift cleanups — extracting subviews, introducing protocol seams, replacing literals with design tokens, adding @MainActor isolation, removing duplication. Use for mechanical improvement with no behavior change. Proves behavior is unchanged by running the tests before and after.