Add agent skills distilled from session usage patterns #2

Merged

adam merged 2 commits from agent-skills into main

2026-07-01 20:56:42 -05:00

adam commented

2026-07-01 17:28:37 -05:00

Owner

Seven project skills under .claude/skills/, mined from ~70 Claude Code session transcripts and the permission allowlist. Each captures the command patterns that actually succeeded against a given API/host, plus the gotchas that cost time in earlier sessions:

forgejo-api — tokens/users, branch+PR via Contents API, LAN-vs-public URL, hard rules for claude creds
op-secrets — service-token read pattern (biometric trap), hermes remote reads, write policy
cloudflare-api — cf-api.sh helper, account-owned token verify quirk, do-not-touch mail records
hindsight-api — proven :2700 REST endpoints, bank list, config-loss gotcha
pangolin-api — the right door (:3003/v1 via docker exec), SSO/git 302 gotcha
homelab-ssh — per-host port/user matrix, 1Password agent, single-line rule
unraid-debug — MCP-first workflow, Dockhand deploy logs, read/write policy

Claude Code picks these up automatically as project skills; other agents (Hermes, CI) can read the plain markdown — index in .claude/skills/README.md.

🤖 Generated with Claude Code

Seven project skills under `.claude/skills/`, mined from ~70 Claude Code session transcripts and the permission allowlist. Each captures the command patterns that actually succeeded against a given API/host, plus the gotchas that cost time in earlier sessions: - **forgejo-api** — tokens/users, branch+PR via Contents API, LAN-vs-public URL, hard rules for claude creds - **op-secrets** — service-token read pattern (biometric trap), hermes remote reads, write policy - **cloudflare-api** — cf-api.sh helper, account-owned token verify quirk, do-not-touch mail records - **hindsight-api** — proven :2700 REST endpoints, bank list, config-loss gotcha - **pangolin-api** — the right door (:3003/v1 via docker exec), SSO/git 302 gotcha - **homelab-ssh** — per-host port/user matrix, 1Password agent, single-line rule - **unraid-debug** — MCP-first workflow, Dockhand deploy logs, read/write policy Claude Code picks these up automatically as project skills; other agents (Hermes, CI) can read the plain markdown — index in `.claude/skills/README.md`. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

adam added 1 commit

2026-07-01 17:28:37 -05:00

docs: add agent skills distilled from session usage patterns 2d8d47a2a0

Seven project skills under .claude/skills/, mined from ~70 Claude Code
transcripts and the permission allowlist: Forgejo API, 1Password reads,
Cloudflare API, Hindsight REST, Pangolin Integration API, SSH matrix,
and Unraid/Dockhand debugging. Each captures the proven command
patterns and the gotchas that cost time in earlier sessions.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

adam added 1 commit

2026-07-01 20:54:17 -05:00

Merge branch 'main' into agent-skills 658ebc4792

adam merged commit 82b8e7d553 into main

2026-07-01 20:56:42 -05:00

adam referenced this pull request from a commit

2026-07-01 20:56:45 -05:00

Merge pull request 'Add agent skills distilled from session usage patterns' (#2) from agent-skills into main

adam referenced this pull request

2026-07-11 19:59:54 -05:00

P2 · Infra-write ergonomics (scoped allow-rules + mode convention) #9

adam referenced this pull request