UNPKG

paqad-ai

Version:

Spec-driven development framework — AI agents that think before they type

github.com/Eliyce/paqad-ai

Eliyce/paqad-ai

53 lines (38 loc) • 2.27 kB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
--- name: request-classifier description: 'Classifies incoming requests into the framework routing model. Use when a new request starts, or when its scope changed enough that earlier routing may no longer hold.' metadata: model_tier: fast cacheable: 'true' output_format: yaml paqad-meta: skill.meta.yaml --- ## What It Does Produces a deterministic request classification across the routing dimensions used by the framework so downstream skill and rule selection can rely on explicit signals instead of guesswork. ## Use This When At the start of a new request or whenever the effective scope changed enough that prior routing assumptions may no longer be safe. ## Inputs - Read the request text, attached plan, and any directly linked issue or spec. - Read the nearest canonical docs only when they clarify stack, module, or risk signals. - Read `references/decision-rules.md` before assigning ambiguous dimension values. ## Procedure 1. Pipe the raw request into `scripts/extract-signals.sh` to get baseline keyword-derived dimensions (workflow, ui/api/db impact, scope, risk hint). 2. Override script defaults only when canonical evidence (linked spec, stack manifest, module map) contradicts them: record the override under Evidence. 3. Use `assets/dimensions.txt` for the allowed value set per dimension. 4. Default to most conservative accurate value when ambiguous; never invent certainty for risk/sensitivity. 5. Format per `assets/output.template.md`; validate with `scripts/lint-output.sh`. ## Output Contract - Match `assets/output.template.md`: `## Classification` (one `key: value` per dimension) and `## Evidence` (bullet list for non-obvious calls). - Output must pass `scripts/lint-output.sh` (exit 0). ## Escalate / Stop Conditions - Ask when stack, workflow, or impact dimensions cannot be inferred safely from the request. - Warn when canonical docs contradict the request about modules, interfaces, or current behavior. - Do not fabricate certainty for risk or sensitivity dimensions. ## Resources - `references/decision-rules.md` - `scripts/extract-signals.sh` - `scripts/lint-output.sh` - `assets/output.template.md` - `assets/dimensions.txt` - `runtime/base/skills/existing-doc-checker/SKILL.md` - `agents/openai.yaml`