User Input
$ARGUMENTS
Audience and tone (interactive mode)
When KISS_AGENT_MODE=interactive (the default), assume the user
has limited technical background and limited domain knowledge
— they may know basics but lack deep expertise in this skill's
area. Run this skill as a guided questionnaire:
- One question at a time. No walls of questions.
- Yes / no first. Phrase so
yes, no, not sure, or skip
is a valid answer.
- Translate jargon, don't strip it. Use the technical term but
always pair it with a plain-English gloss the first time it
appears.
- Choices, not blank fields. When yes/no isn't enough, offer
2-4 lettered options (A/B/C/D) with one-line plain-language
descriptions of the trade-off. Always include "Not sure — pick
a sensible default".
- Always recommend. State the option you would pick and why in
one sentence so the user can reply "yes" / "ok" to accept. Pull
defaults from upstream artefacts (spec, architecture, ADRs,
standards) before asking blank.
- Show, don't ask. When upstream artefacts already imply an
answer, propose it as a pre-filled finding and ask for a
yes / no confirmation rather than asking the user to fill in a
blank.
not sure / skip triggers a sensible default, marked
"(default applied — confirm later)" in the artefact, and a debt
entry in this skill's debt file.
When KISS_AGENT_MODE=auto (or --auto), skip the questionnaire
entirely: apply sensible defaults from upstream artefacts and log
decisions to the parent agent's decision log.
Inputs
.kiss/context.yml{context.paths.docs}/architecture/intake.md (constraints)
Outputs
{context.paths.docs}/research/<topic>.md (one file per topic)
Context Update
Does not mutate .kiss/context.yml.
Handoffs
kiss-adr references research files as evidence in the
"Decision" section.
AI authoring scope
Does: pull current facts via WebSearch/WebFetch (when tools
available); summarise 2–4 candidates in a pros/cons table; record
sources with dates fetched.
Does not: recommend a winner; invent figures; rely on training
data for versioned facts (licensing, pricing, version numbers).
Usage
<SKILL_DIR> = the integration's skills root (e.g. .claude/skills/
for Claude Code, .agents/skills/ for Antigravity / Codex,
.cursor/skills/ for Cursor, .windsurf/workflows/ for Windsurf).
Scripts live at <SKILL_DIR>/<skill-name>/scripts/….
TR_TOPIC=database TR_CANDIDATES="PostgreSQL,MySQL,SQL Server,MongoDB" \
bash <SKILL_DIR>/kiss-tech-research/scripts/bash/draft-research.sh --auto
Answer keys
| Key |
Meaning |
TR_TOPIC |
decision area (database, frontend, identity, …) — required |
TR_CANDIDATES |
comma-separated list — required |
1---2name: kiss-tech-research3description: Researches technology options for a named decision area (frontend framework, database, messaging, identity, etc.) using WebSearch / WebFetch when available. Produces a pros/cons table + typical cost signal per candidate. Records sources. Does not pick the winner. Use when evaluating technology choices, comparing frameworks or databases, or before writing an ADR for a technology decision.4---567## User Input89```text10$ARGUMENTS11```1213## Audience and tone (interactive mode)1415When `KISS_AGENT_MODE=interactive` (the default), assume the user16has **limited technical background and limited domain knowledge**17— they may know basics but lack deep expertise in this skill's18area. Run this skill as a guided questionnaire:1920- **One question at a time.** No walls of questions.21- **Yes / no first.** Phrase so `yes`, `no`, `not sure`, or `skip`22 is a valid answer.23- **Translate jargon, don't strip it.** Use the technical term but24 always pair it with a plain-English gloss the first time it25 appears.26- **Choices, not blank fields.** When yes/no isn't enough, offer27 2-4 lettered options (A/B/C/D) with one-line plain-language28 descriptions of the trade-off. Always include "Not sure — pick29 a sensible default".30- **Always recommend.** State the option you would pick and why in31 one sentence so the user can reply "yes" / "ok" to accept. Pull32 defaults from upstream artefacts (spec, architecture, ADRs,33 standards) before asking blank.34- **Show, don't ask.** When upstream artefacts already imply an35 answer, propose it as a pre-filled finding and ask for a36 yes / no confirmation rather than asking the user to fill in a37 blank.38- **`not sure` / `skip` triggers a sensible default**, marked39 "(default applied — confirm later)" in the artefact, and a debt40 entry in this skill's debt file.4142When `KISS_AGENT_MODE=auto` (or `--auto`), skip the questionnaire43entirely: apply sensible defaults from upstream artefacts and log44decisions to the parent agent's decision log.4546## Inputs4748- `.kiss/context.yml`49- `{context.paths.docs}/architecture/intake.md` (constraints)5051## Outputs5253- `{context.paths.docs}/research/<topic>.md` (one file per topic)5455## Context Update5657Does not mutate `.kiss/context.yml`.5859## Handoffs6061- `kiss-adr` references research files as evidence in the62 "Decision" section.6364## AI authoring scope6566**Does:** pull current facts via WebSearch/WebFetch (when tools67available); summarise 2–4 candidates in a pros/cons table; record68sources with dates fetched.6970**Does not:** recommend a winner; invent figures; rely on training71data for versioned facts (licensing, pricing, version numbers).7273## Usage7475> `<SKILL_DIR>` = the integration's skills root (e.g. `.claude/skills/`76> for Claude Code, `.agents/skills/` for Antigravity / Codex,77> `.cursor/skills/` for Cursor, `.windsurf/workflows/` for Windsurf).78> Scripts live at `<SKILL_DIR>/<skill-name>/scripts/…`.7980```bash81TR_TOPIC=database TR_CANDIDATES="PostgreSQL,MySQL,SQL Server,MongoDB" \82 bash <SKILL_DIR>/kiss-tech-research/scripts/bash/draft-research.sh --auto83```8485### Answer keys8687| Key | Meaning |88|---|---|89| `TR_TOPIC` | decision area (database, frontend, identity, …) — required |90| `TR_CANDIDATES` | comma-separated list — required |
dauquangthanh