subagent-driven-development

A process for carrying out an implementation plan by assigning separate tasks to fresh coding agents and reviewing their work.

In plain words
What is it for?
It helps execute multi-task plans, review each completed task, and perform a final review of the whole branch.
Why use it?
It keeps each task focused and adds checks for both whether the requirements were met and whether the code is well made.

Skill for Claude CodeCodex

Install

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.

agentmods
npx agentmods add skills/missingpackage/nightshift/subagent-driven-development
Any agent
npx skills add MissingPackage/nightshift --skill subagent-driven-development
Clone the repo
git clone --depth 1 https://github.com/MissingPackage/nightshift

Made for: Claude Code, Codex.

Per session 17 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 7,903 The whole file, excluding the scripts and references it only reads on demand.
Security scan E 2 findings. Scan, not verified.
Origin 86% copy Near-identical to another mod in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00017 $0.07903
Opus 5 $0.00009 $0.03952
Sonnet 5 $0.00003 $0.01581
Haiku 4.5 $0.00002 $0.00790

Measured yesterday against content hash 4cfe0582b928, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade E, and why

subagent-driven-development 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 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.

Hidden instructionshighPrompt injection

Directives inside HTML comments, invisible characters or bidirectional overrides are read by the model and not by the person reviewing the file.

<!-- nightshift drift — "decide them" scoped to the authority the goal actually granted. Upstream assumes the running plan is the whole contract; under this harness the goal contract is, and it names an authority boundar

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

delete this plan's workspace (`rm -rf <workspace>`) — the git history is
Origin

This is a copy

86% identical to subagent-driven-development — 23 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.

skills/vendored/subagent-driven-development/SKILL.md · 592 lines

How it starts

The opening of the file, as written. The whole thing — 592 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Subagent-Driven Development

Execute plan by dispatching a fresh implementer subagent per task, a task review (spec compliance + code quality) after each, and a broad whole-branch review at the end.

Why subagents: You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused and succeed at their task. They should never inherit your session's context or history — you construct exactly what they need. This also preserves your own context for coordination work.

Core principle: Fresh subagent per task + task review (spec + quality) + broad final review = high quality, fast iteration

Narration: between tool calls, narrate at most one short line — the ledger and the tool results carry the record.

Continuous execution: Do not pause to check in with your human partner between tasks. Execute all tasks from the plan without stopping. The only reasons to stop are the four named below, or all tasks complete. "Should I continue?" prompts and progress summaries waste their time — they asked you to execute the plan, so execute it.

Rulings, not stalls. A running plan does not wait on a human. Conflicts, ambiguities, plan defects, a cap you would have asked to exceed — decide them. The spec is the binding authority, the plan is its argument, and your judgment settles what neither answers. Record every decision in the ledger as Ruling: <what you decided> — <why> — <what it costs if wrong>, and keep going. A wrong ruling costs rework your human partner can see and undo; a session parked on a question costs their whole day and buys nothing.

Four things stop you, and only these: an irreversible or destructive operation; a security-sensitive action; a side effect outside this worktree that norms say you ask about first (a merge, a push to a shared branch, a publish); and a plan so broken that every path forward is a guess. For those, stop and ask.

Scoping under a goal contract. When this skill runs inside a goal (.harness/goals/<slug>/), "decide them" means decide within the authority the goal granted — not beyond it. The escalation ladder in ORCHESTRATION.md §4 is the binding version, and it is stricter than the four stops above:

Situation What to do
Ambiguity with a safe default; out-of-scope finding Rule it. Record Ruling: … — … — …, mark [ASSUMED] in the journal, keep going.
Authority edge the goal did not grant (merge, protected path, new dependency, schema or public-API change) Docket it, block the phase. Take the next independent phase if one exists.
Second verifier failure on the same phase; budget at or past its cap Docket it, block the phase. A retry needs a new hypothesis — never the same plan twice.
Secrets or destructive operations; contradictory directives Pause the goal and notify.

What upstream is right about survives: a session parked on a question the agent had the authority to answer costs the user their whole day and buys nothing. The correction is only about which questions those are — the goal contract decides that, not the agent.

Read the full file on GitHub · 592 lines

Files

What ships with it

6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

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.

  1. yesterday First seen · 592 lines · 17 tokens per session scan E 4cfe0582b928

Subscribe to this mod's changes

subagent-driven-development is a skill published in the GitHub repository MissingPackage/nightshift (2 stars, last pushed 15d ago), licensed MIT. It adds 17 tokens to every session and 7,903 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it E with 2 findings (hidden instructions, recursive force delete). It is 86% identical to subagent-driven-development, differing in 23 lines, and is treated as a copy.

Related

Other skills, from other repositories

baton-setup

One-time setup and health check for fable-baton. Use when the user asks to set up, configure, verify, or troubleshoot fable-baton - sets the default model to "best" (Fable 5 with Opus fallback) in /.claude/settings.json and verifies the plugin is fully installed.

realgarit/fable-baton · 69 tokens

licensing-tiers-data-governance

Implement subscription tiers with field-level access control, feature gating, rate limiting, and compliance tracking. Design data governance systems that enforce different access levels, retention policies, and regulatory requirements based on user subscription tier (Free, Pro, Enterprise).

sunnypatneedi/claude-starter-kit · 56 tokens

data-infrastructure-at-scale

Build data infrastructure that scales from prototype to production. Use when architecting data pipelines, choosing data stores, planning for high throughput, or migrating to distributed systems. Covers caching, replication, sharding, message queues, and data lake architecture.

sunnypatneedi/claude-starter-kit · 56 tokens

data-provenance

Track data lineage and provenance from source to consumption. Use when auditing data flows, debugging data quality issues, ensuring compliance (GDPR, SOX), or understanding data dependencies. Covers lineage tracking, impact analysis, data catalogs, and metadata management.

sunnypatneedi/claude-starter-kit · 56 tokens

multi-source-data-conflation

Merge and reconcile data from multiple sources into a unified view. Use when integrating APIs, consolidating databases, building data warehouses, or creating master data. Covers entity resolution, conflict resolution, data quality, and real-time vs batch conflation.

sunnypatneedi/claude-starter-kit · 57 tokens

software-architecture

Design scalable software systems with proven architectural patterns (MVC, microservices, event-driven), SOLID principles, system design trade-offs, and architectural decision records (ADRs).

sunnypatneedi/claude-starter-kit · 38 tokens