Orient
Start issue-driven work by pulling GitHub context, inferring feat/fix/chore/docs branch prefixes, and surfacing ADRs before coding.
Overview
orient is an agent skill most often used in Build (also Ship, Operate) that fetches issue context, detects task type, maps branch prefixes, and presents a brief before deeper implementation work.
Install
npx skills add https://github.com/camacho/ai-skills --skill orientWhat is this skill?
- EXPLORE phase gate: all steps must finish before proceeding to Isolate
- Fetches issue JSON via gh when an issue number is provided
- Maps labels and keywords to branch prefixes fix/, feat/, chore/, docs/
- Scans ai-workspace/decisions ADRs and flags conflicts with the task
- Resumes interrupted work when .branch-context.md exists at repo root
- Five-step EXPLORE gate including issue fetch, task-type table, ADR scan, and interrupted-work check
Adoption & trust: 630 installs on skills.sh; 1 GitHub stars; 2/3 security scanners passed (skills.sh audits).
What problem does it solve?
You opened a repo on a vague ticket without a branch prefix, ADR context, or memory of half-finished work from last session.
Who is it for?
Solo builders on GitHub-issue workflows who keep ADRs in-repo and want a mandatory EXPLORE gate before coding.
Skip if: Greenfield spikes with no issues tracker, or teams that do not use git branches named feat/fix/chore/docs.
When should I use this skill?
EXPLORE checkpoint when starting issue-driven work and before proceeding to Isolate.
What do I get? / Deliverables
You get a structured orient brief with task type, suggested branch prefix, relevant ADRs (including conflicts), and resume notes so you can proceed to Isolate with shared context.
- Orient brief with task type and branch prefix
- ADR relevance and conflict callouts
Recommended Skills
Journey fit
Spans multiple journey phases - primary shelf plus alternate fits below.
Orient sits at Build/pm as the canonical entry ritual when turning a ticket into a branch-ready brief—the first structured step in the camacho EXPLORE gate. PM subphase fits task framing, branch naming, and architecture decision lookups rather than frontend implementation or shipping automation.
Where it fits
Pull issue #42 labels to propose feat/ and surface ADR-003 before building the feature.
Classify a bug report as fix/ and attach ADR notes before you isolate the regression.
Read .branch-context.md to resume a chore/ branch after an agent session ended mid-task.
Detect docs/ prefix from readme-related keywords when no bug or feature labels exist.
How it compares
Use instead of improvising branch names in chat when you want issue labels, ADRs, and resume state in one gate.
Common Questions / FAQ
Who is orient for?
Solo and indie developers using GitHub issues and a phased agent workflow who need a consistent EXPLORE briefing before isolation and implementation.
When should I use orient?
Use it at the start of Build/pm when picking up an issue; in Ship when scoping a fix/ branch bug; in Operate/iterate when resuming from .branch-context.md after interrupted work.
Is orient safe to install?
It instructs gh and git reads against your repo; review the Security Audits panel on this Prism page and avoid exposing secrets in issue bodies the skill will fetch.
SKILL.md
READMESKILL.md - Orient
Phase gate: EXPLORE checkpoint. This skill MUST complete all steps before proceeding to Isolate. ## Steps 1. **Fetch issue context** (if issue number provided): ```bash gh issue view <number> --json title,body,labels,assignees ``` If no issue number provided, ask the user to describe the task. If `gh` is not available (remote session without GH_TOKEN), skip issue fetch and proceed with manual description. 2. **Auto-detect task type** from labels or description: | Label / keyword | Type | Branch prefix | |----------------|------|---------------| | `bug`, error, crash, broken, fix | Bugfix | `fix/` | | `enhancement`, `feature`, add, want, should, new | Feature | `feat/` | | `chore`, `config`, deps, ci, tooling | Chore | `chore/` | | `docs`, documentation, readme | Docs | `docs/` | | Default | Feature | `feat/` | 3. **Scan for relevant ADRs**: ```bash ls "$(git rev-parse --show-toplevel)/ai-workspace/decisions/"*.md 2>/dev/null | head -20 ``` Check ADR titles/filenames for keywords matching the issue title/body. If matches found, read the relevant ADR and surface it. If ADR scan finds contradictions, surface explicitly: "This task may conflict with ADR-NNN" 4. **Check for interrupted work**: If `"$(git rev-parse --show-toplevel)/.branch-context.md"` exists, read it and include in the brief (resuming work). 5. **Read recent memory**: ```bash tail -20 "$(git rev-parse --show-toplevel)/ai-workspace/MEMORY.md" 2>/dev/null ``` 6. **Present brief**: ``` ## Orient Brief **Task**: [title or description] **Type**: [bugfix/feature/chore/docs] → branch prefix: [fix/feat/chore/docs]/ **Suggested branch**: [prefix]/[kebab-case-name] **Relevant ADRs**: [list or "none found"] **Resuming?**: [yes — from .branch-context.md / no] **Files likely affected**: [educated guess from issue description + ADR context] ``` 7. **Auto-route** using the one-sentence rule: - Can the entire diff for this task be described in ONE sentence? → **Direct to Isolate** (skip Design/Review) - Otherwise → **Plan first** (proceed to Design after Isolate) Present the routing decision. The agent auto-decides — no human input required. If human is present, they can override. ## Edge Cases - No issue number + no user description → ask for a description - `gh` not available → skip issue fetch, proceed with manual description - No ADRs directory → skip ADR scan