Context-Load

Reads project documentation and builds a compressed snapshot that later phases (issue-craft, pr-context) consume. The snapshot is never a copy of file contents — it is a structured index of paths + key facts.

Reading Order

Read in order, skip missing files:

1. AGENTS.md — workflow, conventions, file reference table
2. ARCHITECTURE.md or docs/ARCHITECTURE.md — architecture decisions
3. docs/DEVELOPMENT.md — dev guide, build/test commands
4. docs/ACCEPTANCE_CRITERIA.md (or equivalent) — AC patterns by issue type
5. .github/ISSUE_TEMPLATE/* — issue template fields and auto-labels
6. .github/PULL_REQUEST_TEMPLATE/* — PR template required sections
7. .github/workflows/* — CI check names (extract job names, not full YAML)
8. README.md — project snapshot (test counts, file counts, stack)

Conventions Extraction

After the reading order, scan the following files for imperative rules and framework-specific conventions that affect issue/PR crafting:

• AGENTS.md — scan for imperative rules (keywords: "must", "always", "never", "MUST", "CRITICAL"). Extract the rules that constrain code or workflow decisions.
• CONTRIBUTING.md, DEVELOPMENT.md, ADEV.md — scan for conventions (naming, structure, patterns). Skip files that are absent.
• Code review checklists (if present, e.g. .github/CHECKLIST.md, docs/CODE_REVIEW.md) — extract enforcement rules that reviewers apply.
• Framework-specific API rules — e.g. "use OnClientEvent, not .Event" for RemoteEvents.
• Lifecycle rules — e.g. "disconnect all :Connect() in KnitDestroy".
• Testing conventions — e.g. "mock Players:GetPlayerByUserId → nil for no-op replication".
• Rollout/deployment conventions — e.g. "3-stage rollout (Hidden → Integrated → Cleanup) for new features".

Keep the extracted conventions compact: max 10 bullets, only the rules that affect issue/PR crafting. Drop generic advice ("write clean code") and keep only enforceable, framework-specific rules.

Snapshot Schema

Write to the progress file (.slim/acf/<slug>.md):

## Context Snapshot

### Architecture
- Stack: [languages, frameworks, build tools — one line]
- Key dirs: [src/, scripts/, docs/ — paths only]
- Architecture doc: [path or "none"]

### Challenge
- Current branch: [name]
- Open issues: [count] (from stack-audit)
- Open PRs: [count] (from stack-audit)

### Templates
- Issue templates: [list names + auto-labels]
- PR template: [path, required sections]

### Tests
- Test commands: [exact commands, one per line]
- Test count: [N tests across M suites]
- Coverage tool: [command or "none"]

### CI Checks
- Check names: [list job names from workflows]
- Required checks: [which ones gate merge]

### Conventions
- API: `OnClientEvent`/`OnServerEvent` for RemoteEvents (NOT `.Event`)
- Lifecycle: all `:Connect()` stored + disconnected in `KnitDestroy`
- Constants: no magic numbers — all in `Constants.luau` as UPPER_SNAKE_CASE
- Testing: mock `Players:GetPlayerByUserId` → `nil` for no-op replication
- Rollout: 3-stage (Hidden → Integrated → Cleanup) for new features
- Commit format: [conventional commits or custom]
- Branch naming: [prefix rules]

Compression Rules

1. Paths, not contents — src/shared/Constants.luau not the file's code
2. Command strings, not configs — lune run scripts/test_core.luau not the test file
3. Check names, not YAML — test (lune) not the full workflow
4. Counts, not lists — "26 services" not a list of all 26 service names
5. Max 5 bullets per section — if more, summarize

Output

The snapshot is consumed by:

• stack-audit (phase 2) — uses test commands and CI check names
• issue-craft (phase 3) — uses AC patterns, templates, conventions
• pr-context (phase 4) — uses test commands, CI check names, PR template

Empty State Handling

If key project files are missing, the snapshot still builds — record "none" or "not found" for the missing section instead of skipping it:

• No ARCHITECTURE.md — set Architecture doc: none. Infer the stack from package.json, Cargo.toml, go.mod, or file extensions. Note in the snapshot: "Architecture doc missing — stack inferred from manifest files."
• No .github/workflows/ — set CI Checks: none. Note: "No CI workflows detected — AC should reference local test commands only."
• No .github/ISSUE_TEMPLATE/ — set Issue templates: none. Use the default issue structure from issue-craft (this skill's body template).
• No .github/PULL_REQUEST_TEMPLATE/ — set PR template: none. Use the default PR body structure from pr-context.
• No README.md — set Stack: inferred from file extensions. Note: "No README — stack inferred from source files."
• No AGENTS.md — skip conventions extraction, set Conventions: none.

The snapshot is always produced, even for a bare repo with only source files.

Error Handling

• gh CLI not installed: context-load does not require gh — it reads files from the local filesystem. If gh is needed for issue/PR counts, note "gh not available — counts deferred to stack-audit" and proceed.
• gh not authenticated: same as above — context-load works without gh. Stack-audit will handle auth errors.
• GitHub API rate limits: not applicable to context-load (no API calls). If gh is used for supplementary data and rate-limited, skip that data and note it in the snapshot.
• Required files/directories don't exist: see Empty State Handling above. Missing files are expected — record "none" and continue.
• Commands fail: if a file read fails (permissions, encoding), skip the file and note it: "[file]: read error — skipped". Do not abort the snapshot.

Example

Filled-in snapshot for a sample Luau game project:

## Context Snapshot

### Architecture
- Stack: Luau, Knit, Lune build tool
- Key dirs: src/server/, src/client/, src/shared/, scripts/, docs/
- Architecture doc: docs/ARCHITECTURE.md

### Challenge
- Current branch: fix/flashlight-fireserver
- Open issues: 7 (from stack-audit)
- Open PRs: 2 (from stack-audit)

### Templates
- Issue templates: bug-report.yml (auto-labels: bug, area:backend), feature-request.yml (auto-labels: enhancement)
- PR template: .github/PULL_REQUEST_TEMPLATE.md (required: Summary, Closes, AC)

### Tests
- Test commands: lune run scripts/test_core.luau
- Test count: 26 tests across 3 suites
- Coverage tool: none

### CI Checks
- Check names: test (lune), lint (selene), type-check (luau-analyze)
- Required checks: test (lune), type-check (luau-analyze)

### Conventions
- API: OnClientEvent/OnServerEvent for RemoteEvents (NOT .Event)
- Lifecycle: all :Connect() stored + disconnected in KnitDestroy
- Constants: no magic numbers — all in Constants.luau as UPPER_SNAKE_CASE
- Testing: mock Players:GetPlayerByUserId → nil for no-op replication
- Rollout: 3-stage (Hidden → Integrated → Cleanup) for new features
- Commit format: conventional commits
- Branch naming: fix/, feature/, docs/