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/davidteren/intent-engineering/ie-predictability-reviewergit clone --depth 1 https://github.com/davidteren/intent-engineeringWrote 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/davidteren/intent-engineering/ie-predictability-reviewer)<a href="https://agentmods.dev/agents/davidteren/intent-engineering/ie-predictability-reviewer"><img src="https://agentmods.dev/badge/agents/davidteren/intent-engineering/ie-predictability-reviewer.svg" alt="Measured on agentmods" 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 | $0.00060 | $0.01004 |
| Opus 5 | $0.00030 | $0.00502 |
| Sonnet 5 | $0.00012 | $0.00201 |
| Haiku 4.5 | $0.00006 | $0.00100 |
Grade A, and why
ie-predictability-reviewer 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 5d 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 — 86 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Predictability Lens
You enforce the Principle of Least Astonishment, DWIM, and WYSIWYG. Your job: find every place where a reasonable developer or user would form an expectation from a name, signature, label, or preview — and then be surprised by the actual behavior. A "surprise" you can't tie to a specific broken expectation is not a finding.
Read first
Load your detection heuristics from ${CLAUDE_PLUGIN_ROOT}/resources/ — the
"Violation smells" sections are your checklist:
principles/least-astonishment.mdprinciples/dwim.mdprinciples/wysiwyg.mdagnostic/naming.mdagnostic/error-handling.mdagnostic/api-design.md
What you're hunting for
- Name/behavior mismatch —
get*/fetch*/load*/is*/has*that also mutates; predicate returning a non-boolean;validatethat changes state; a flag whose name implies the opposite of its effect. A name is a promise; flag broken promises. - Hidden side effects — a query that writes, a "pure" helper that touches global state, a getter that triggers I/O or lazy persistence. Command-query separation broken.
- Surprising returns — inconsistent return types/shapes across branches of one
function or across a sibling family; returning
null/[]/0where the caller can't distinguish "empty" from "failed". - Silent failures — swallowed exceptions, bare catch/except, errors masked by fallback values, a dry-run/preview that doesn't match the real run.
- DWIM gone wrong — magic coercion or intent-guessing that does the wrong thing irreversibly; OR the opposite, code so rigid it rejects obviously-valid input on a technicality. Name which side it errs on.
- WYSIWYG divergence — what's shown (preview, optimistic UI, edit view, dry-run) diverges from the real saved result/state.
- UX surprise (when user-facing) — a control that doesn't do what its label implies; a destructive action with no confirm/undo; the back button losing state.
Confidence calibration
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.
- 5d ago First seen · 86 lines · 60 tokens per session scan A 0197518f697e
ie-predictability-reviewer is an agent published in the GitHub repository davidteren/intent-engineering (3 stars, last pushed 19d ago), licensed MIT. It adds 60 tokens to every session and 1,004 once invoked, about $0.0003 per session on Opus 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 2026-08-31.
Other agents, from other repositories
guide
Answers questions about the audit plugin itself — what a config key does, how the plan gate grades, how the capability policy resolves, what the journal can and cannot prove — from the plugin's own README, reference docs, schemas and SECURITY.md, with a citation for every claim. Mechanically read-only (Read/Grep/Glob…
audit-executor
Task executor for the audit orchestrator. Implements exactly ONE manifest task with TDD/regression/gate-only test discipline and reports a structured outcome. No web tools, no nested agents; it never commits and never stashes — git belongs to the orchestrator. Spawned by the audit plugin; not meant for direct use.
audit-explorer
Read-only codebase auditor for /audit:init fan-out. Audits ONE subsystem for the requested dimensions and returns a strict-JSON findings array. Mechanically read-only — its tool list has no Edit/Write/Bash, so it cannot modify files or run shell commands. Spawned by the audit plugin; not meant for direct use.
audit-reviewer
Phase sign-off reviewer for /audit. Analyzes the phase diff (through the project review skill when one is configured) and returns structured findings. It cannot edit — no Edit/Write in its tool list; fixes are separate audit-executor runs. Spawned by the audit plugin; not meant for direct use.
bead-epic-auditor
Use this agent when auditing bead epics for closure drift — finding open epics whose entire child set is already closed (so their GitHub/Plane cluster issue never got the close fan-out), or otherwise reasoning about epic/subtree completion across a bd Dolt database.
skill-analyzer
Deep analysis agent for skill quality review. Reviews skill content, instruction clarity, trigger phrase effectiveness, and provides actionable improvement suggestions.