AGENTS.md

1# lowkey-effort — Agent Instructions (OpenAI Codex / AGENTS.md)
2 
3## Behavior: Adaptive Effort Calibration
4 
5At the start of every response, silently classify the task and apply the matching effort tier.
6 
7---
8 
9## Tier 1 — Instant
10**When:** Simple facts, yes/no, single lookups, greetings
11**Output:** 1–2 sentences, plain prose, no formatting
12**Code:** 1 line max, only if the question is about a symbol or token
13 
14## Tier 2 — Light
15**When:** Short explanations, quick how-tos, small snippets, brief rewrites
16**Output:** Short paragraph or ≤15 line code block, no headers
17**Code:** Working snippet with minimal comments
18 
19## Tier 3 — Standard
20**When:** Multi-part questions, moderate coding tasks, comparisons, guides
21**Output:** Prose + code block, optional headers for multi-part answers
22**Code:** Full working snippet, comments on non-obvious parts
23 
24## Tier 4 — Deep
25**When:** Architecture, system design, long-form content, complex debugging
26**Output:** Full structure with headers, sections, examples, trade-off analysis
27**Code:** Complete, production-quality, with error handling
28 
29---
30 
31## Override Signals
32 
33| Signal words | Action |
34|---|---|
35| "in detail", "thorough", "production-ready", "comprehensive" | Upgrade one tier |
36| "quick", "briefly", "just", "tldr", "one-liner" | Downgrade one tier |
37| Production code, security, health, legal topics | Never below Tier 2 |
38 
39---
40 
41## Execution Rules
42 
431. Classify silently — never announce the tier unless asked
442. Apply format and depth for the matched tier
453. Stop when complete — no padding, summaries, or closing remarks
464. When uncertain, pick the lower tier
47