Argument Parsing

Standardized rules for how all GHS skills parse $ARGUMENTS --- the raw string passed when a user invokes a skill.

Invocation Example

/ghs-backlog-fix owner/repo --item tier-1--license --dry-run

Repo Detection

Skills auto-detect the target repository using this priority:

Priority	Source	Pattern
1	First positional arg	owner/repo
2	Full GitHub URL	https://github.com/owner/repo --- extract owner/repo
3	Auto-detect	gh repo view --json nameWithOwner -q .nameWithOwner
4	Prompt	Ask user if detection fails

URL normalization:

• Strip https://github.com/ prefix
• Strip trailing /, .git, /issues/..., /pulls/...
• Result must match ^[^/]+/[^/]+$

Flag Extraction

All flags use -- prefix. No short flags (e.g., -d) are supported.

Boolean Flags

Flag	Behavior
--dry-run	Show what would happen without executing
--all	Process all eligible items instead of prompting to select
--auto	Skip interactive confirmations
--resume	Resume from last persisted state

Value Flags

Flag	Value Type	Example
--item <slug>	string	--item tier-1--license
--tier <N>	integer (1--3)	--tier 2
--pr <number>	integer	--pr 42
--budget <N>	integer	--budget 5
--bump <level>	major|minor|patch	--bump minor
--stages <list>	comma-separated	--stages pull,scan,fix

Issue/PR Number Parsing

Input Format	Result
#42	issue/PR number 42
bare 42 (after repo arg)	issue/PR number 42
owner/repo#42	repo = owner/repo, number = 42
https://github.com/owner/repo/issues/42	repo = owner/repo, number = 42
https://github.com/owner/repo/pull/42	repo = owner/repo, number = 42

Disambiguation: If a bare integer appears without explicit --pr or --item, treat as an issue/PR number only when the skill context expects one (e.g., ghs-issue-implement, ghs-review-pr).

Item Slug Normalization

Project item slugs use kebab-case with double-dash tier prefix: tier-1--license, ci-workflow-health.

Input	Normalized Match
tier-1--license	exact match
license	partial match --- tier-1--license (first match wins)
ci-workflow	partial match --- ci-workflow-health
TIER-1--LICENSE	case-insensitive match

Partial match rule: Match any slug that contains the input as a substring (case-insensitive). If multiple slugs match, prefer the one with the shortest total length.

Defaults When No Arguments

Scenario	Behavior
No args at all	Auto-detect repo from gh repo view; use skill defaults for all flags
Repo not detectable	Prompt: "Which repo? (owner/repo)"
Flag omitted	Use skill-defined default (documented per skill)
Ambiguous partial slug	List matches and prompt user to select

Common Flag Patterns

Universally available flags across skills:

Flag	Type	Skills That Use It	Behavior
--dry-run	boolean	backlog-fix, backlog-sync, merge-prs, orchestrate, release	Show what would happen without executing
--all	boolean	backlog-fix, issue-triage, merge-prs	Process all items instead of selecting
--auto	boolean	issue-triage, orchestrate	Skip interactive confirmations
--resume	boolean	orchestrate	Resume from last persisted state

Parsing Order

Recommended parse order from $ARGUMENTS:

1. Extract owner/repo   --- first positional token matching owner/repo or full URL
2. Extract issue/PR #   --- token matching #N, bare N (context-dependent), or URL
3. Extract flags        --- all --prefixed tokens and their values
4. Remaining tokens     --- free-form identifiers: slugs, descriptions, search terms

Priority conflicts: Explicit --pr or --item flags override positional detection.