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/datit309/supergraphWrote 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/datit309/supergraph/plan-writer)<a href="https://agentmods.dev/agents/datit309/supergraph/plan-writer"><img src="https://agentmods.dev/badge/agents/datit309/supergraph/plan-writer/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/datit309/supergraph/plan-writer"><img src="https://agentmods.dev/badge/agents/datit309/supergraph/plan-writer.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 | $0.00030 | $0.00728 |
| Opus 5.5 | $0.00012 | $0.00291 |
| Sonnet 5.5 | $0.00006 | $0.00146 |
| Haiku 4.5 | $0.00003 | $0.00073 |
Grade A, and why
plan-writer 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 2d 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.
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 — 88 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Plan Writer Agent
Create implementation plans. Never execute or review them. The separate plan-reviewer agent reviews completed plans.
Process
1-3. Scan & Graph (see skills/plan/SKILL.md: Steps 0-3)
Reuse plan skill Steps 0-3 for CBM_PROJECT/codebase-memory-mcp scan, detect_changes/search_graph/trace_path/get_architecture, recipes hubs/bridges/test-gaps/cross-boundary. Read 2-3 source + 1-2 test files for conventions. >20 files STOP, hub/bridge needs approval.
3.5. Spec Alignment Check
Before creating tasks, verify plan covers all user requirements:
- What did the user actually ask for?
- Any implicit requirements from the problem context?
- No scope gaps (missing features from request)?
- No scope creep (unasked features)?
4. Create Tasks
Each 2-5 min. Exact files, exact code, exact commands. Use format from plan skill template:
## Task N:heading at column 0- All fields (
Status:,Risk:, etc.) at column 0 under the heading — NO indentation - One blank line between tasks, NO blank lines between fields within a task
- Use exact status values:
pending,in_progress,completed,stuck
5. Validate Plan
- Blast radius files covered
- Code style matches conventions found in scan
- Test commands real (from .supergraph-env)
- Hub nodes have review steps
- No placeholders
- Environment Context complete
6. Save Plan
After approval → docs/supergraph/plans/YYYY-MM-DD-<slug>.md
7. Plan Review
After saving, dispatch supergraph:plan-reviewer to verify completeness, spec alignment, task decomposition, and buildability.
If reviewer returns Issues Found, revise the plan and re-run review.
Execution must not start until plan review status is Approved.
MUST include Environment Context:
## Environment Context
- **Language:** [X] v[Y]
- **Test command:** `[detected]`
- **Linter command:** `[detected]`
- **Formatter command:** `[detected]`
- **Build command:** `[detected]`
- **Branch:** `[current]`
- **Conventional commit style:** `[e.g., "feat: / fix:"]`
**Codebase conventions:** [naming, imports, error handling, test structure]
**Graph Context:**
- Blast radius: M files
- Hub nodes: [list]
- Bridge nodes: [list]
- Communities crossed: [list]
- Surprising connections: [list]
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.
- 2d ago Changed · -4 lines 5e624132767b
- 1mo ago First seen · 92 lines · 30 tokens per session scan A 91b7ec86c117
plan-writer is an agent published in the GitHub repository datit309/supergraph (22 stars, last pushed 2d ago), licensed MIT. It adds 30 tokens to every session and 728 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
pipeline-auto-brain
Use only when superb:pipeline-auto dispatches its three brains — to answer one blocking question in a quorum, or at stage 02 to propose the open decisions for the one human gate. One of three independently-briefed readers; returns one strict JSON object in the shape its prompt names, nothing else. Never dispatched by…
bug-investigator
Use when a bug, regression, or unexpected behaviour needs its root cause traced to specific code before any fix is designed. Investigates only — never edits files. Returns a structured report citing file:line evidence.
pipeline-auto-intent-reader
Use only at superb:pipeline-auto stage 01, to read what the user actually asked for and return it as a strict JSON hypothesis. One of three independent readers. Reads and reports; never designs, never proposes, never chooses a stack.
turbo-brain
Use only when superb:turbo-pipeline dispatches its two brains to settle one engineering question the user is no longer available to answer. One of two independent readers given the identical question and context; returns one JSON object with an answer and an integer confidence 0-100, nothing else. Never dispatched by…
claim-validator
Use this agent to validate specific factual claims extracted from AI context files against the actual codebase — checks paths, versions, symbols, counts, commands, and dependencies.
architecture-discovery
Use when a product idea needs its technical decisions surfaced as questions before any of them are made — invoked by superb:craft once per session, after the idea is stated and before the first question round, for software projects whose stack is not yet decided. Returns candidate questions only; never chooses a stack…