sgtlaugh/skills

░▒▓ skills ▓▒░
liar ahead. try grep.
Claude Code skills that make the agent prove it.
Evidence or nothing: every verdict cites a grep, a query or a file:line, or says Can't-determine.
Fresh eyes: reviews run in a separate context, so the author's reasoning can't wave its own work through.
## Install
1. Clone the repo and run the installer:
bash
git clone https://github.com/sgtlaugh/skills.git ~/Github/skills
~/Github/skills/install.sh
2. Read its checklist. Skills with nothing in their Needs column work now; each ⚠ line says what the rest are missing:
text
Config → ~/.config/claude-skills
⚠ credentials.json 2 of 10 keys set
⚠ services.md still the template
Environment
⚠ gh not logged in - run gh auth login → /aristarchus, /nicepr, /triage, /sentry-fix
3. Set up only what your skills need:
Jira, Sentry, alert keys: add them to ~/.config/claude-skills/credentials.json, which the installer created (mode 600). Where to get each one is under Config files and credential keys below.
json
{ "jiraBaseUrl": "https://acme.atlassian.net", "jiraEmail": "you@acme.com", "jiraApiToken": " " }
gh: gh auth login
$WORKSPACE: export WORKSPACE=~/code in your shell profile, the folder holding your clones
services.md: fill in the template in the same folder: your repos, log search and test commands
4. Re-run install.sh until the checklist is green, then restart Claude Code and type /. Re-run it after every pull.
Every skill needs Claude Code, jq and curl 7.76+, on Linux or macOS with bash 3.2+. Common usage is in CHEATSHEET.md; every flag is in each skill's SKILL.md.
Config files and credential keys
Up to four files live in ~/.config/claude-skills/. install.sh creates the first two if they're missing, never overwrites them, and backs up anything in its way to .bak. Skills that need none of them work right after install.
| File | Created by install.sh | Used by | Holds |
| credentials.json | yes (mode 600) | the API skills | API tokens (keys below) |
| services.md | yes | /sentry, /sentry-fix, /triage, /firewatch, /aristarchus | Your systems: repos, log search, alert patterns, test commands |
| voice.md | no | /humanize | Your voice for GitHub text |
| work.json | no | /nicepr | {"reviewers": [...]} for --create |
credentials.json lives elsewhere if CLAUDESKILLSCREDENTIALS points there.
services.md says which repo each Sentry project lives in, your log search skill, your alert patterns and how to run each repo's tests. It starts as a template whose values read ; skills treat those as missing and ask, so fill in the sections your skills use.
voice.md holds how you suggest, push back and approve.
work.json's reviewers are asked to review a PR /nicepr --create opens, unless it's a draft.
credentials.json keys
| Key | Used by | Where to get it |
| datadogPAT | /firewatch | Datadog → Personal Access Tokens |
| datadogSite | /firewatch | Your Datadog site, e.g. datadoghq.com |
| jenkinsApiToken | /firewatch | Jenkins → your user → Configure → API Token |
| jenkinsUser | /firewatch | Your Jenkins username |
| jiraApiToken | /jira, /gadfly, /forge, /duck, /nicepr, /aristarchus, /triage, /firewatch | Atlassian API tokens |
| jiraBaseUrl | /jira, /gadfly, /forge, /duck, /nicepr, /aristarchus, /triage, /firewatch | Your Atlassian site, e.g. https://acme.atlassian.net |