Ralph Specum Start

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

.claude/skills/Command Development

.claude/skills/Hook Development

.claude/skills/MCP Integration

.claude/skills/Plugin Settings

.claude/skills/Plugin Structure

Audit 0.4.1 · source sha a73b124937d9 · 18 Oct 2026Report an issue or request a re-scan

• Publishertzachbon GitHub userOwner of the repository this skill is read from
• Source foldertzachbon/smart-ralph/plugins/ralph-specum-codex/skills/ralph-specum-startThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdicta73b124937d9A new audit runs when the source changes
• Repository statusNot archivedLast push 32 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: This skill should be used only when the user explicitly asks to use `$ralph-specum-start`, or explicitly asks Ralph Specum in Codex to start or resume a spec.

---
name: ralph-specum-start
description: This skill should be used only when the user explicitly asks to use `$ralph-specum-start`, or explicitly asks Ralph Specum in Codex to start or resume a spec.
metadata:
  surface: helper
  action: start
---

# Ralph Specum Start

Use this for the `start` and `new` entrypoints.

Derive `RALPH_CODEX_PLUGIN_ROOT` from this loaded skill by resolving two parent directories from the `SKILL.md` directory. Never derive it from the project working directory.

## Contract

- Read `.claude/ralph-specum.local.md` when present
- Default specs root is `./specs`
- Keep `.current-spec` in the default specs root
- Keep the standard Ralph files stable
- Merge `.ralph-state.json`. Do not replace the full object

## Action

1. Parse explicit name, goal, exact `--quick` or exact `--interactive`, `--resume <prototype-id>`, commit flags, optional specs root, and optional `--tasks-size fine|coarse`. Reject both mode flags together, `-q`, variants, and natural-language substitutes.
2. Classify new versus resume intent, then resolve the classified target by explicit path, exact name, or `.current-spec`. A new-spec request does not inherit or recover the existing current spec.
3. If the same name exists in multiple configured roots, stop and require a full path.
4. Check active epic context from `specs/.current-epic` when no explicit spec was chosen.
5. For large or cross-cutting goals, route to triage instead of forcing a single spec.
6. `new` is an alias here. Create the spec directory if needed.
7. Initialize or merge state with:
   - `source: "spec"`
   - `name`
   - exact `goal` text used by the context gate
   - `basePath`
   - `phase: "research"`
   - `taskIndex: 0`
   - `totalTasks: 0`
   - `taskIteration: 1`
   - `maxTaskIterations: settings default or 5`
   - `globalIteration: 1`
   - `maxGlobalIterations: 100`
   - `commitSpec: settings auto_commit_spec or true`
   - `relatedSpecs: []`
   - `awaitingApproval: false` while the goal interview is active
   - preserve or set `quickMode`
   - preserve or set `granularity` when `--tasks-size` was supplied
   - preserve or set `epicName` when starting from an epic suggestion
8. Update `.current-spec`.
9. Write `.progress.md` with goal, current phase, next step, blockers, learnings, and skill discovery results.
10. On resume, prefer `tasks.md` and present files over stale state when they disagree.
11. Run `phase_gate.py mode` through `"$RALPH_CODEX_PLUGIN_ROOT/scripts/phase_gate.py"` with `STATE` and the exact supplied mode flag. No flag normalizes invalid legacy quick state to interactive.
12. Run skill discovery pass 1 against the goal. Collect plugin skills, project `.agents/skills`, project `.claude/skills`, and the current Codex harness catalog. Always select explicitly named skills and record shadowed duplicates.
13. Load `"$RALPH_CODEX_PLUGIN_ROOT/skills/interview-framework-codex/SKILL.md"`, its required algorithm and domain-modeling references, and every selected domain contract in both interactive and quick mode. In interactive mode, follow that algorithm for the start goal territory: outcome and observable success, material scope boundaries, critical constraints, and viable high-level approach. Ask no spec-path, branch, task-size, or discoverable question.
14. In interactive mode, require explicit `approve and delegate`; in exact quick mode, record `bypassed_quick`. In both modes, run `phase_gate.py check-delegation` with the current loaded-manifest identity before creating a `research-analyst` child. Apply the shared hard-transition invariant before fresh or resumed research dispatch. A failed `check-delegation` in either mode stops this invocation before phase transition, child dispatch, or target-artifact write; after a normal-mode failure, only the next explicit invocation creates a fresh manifest/interview identity. After an exact `--quick` delegation failure, the next explicit invocation reruns discovery, records a fresh `phaseSkillLoad` and interview identity, and does not reuse a terminal `bypassed_quick` interview or its discovery revision. Only a matching in-progress `collecting` or `awaiting_confirmation` interview is resumable. Exact `--quick` retains its existing question-and-approval bypass and discovery, manifest, delegation, and writer checks.
15. Keep the existing child packet, identity tuple, and receipt behavior unchanged: pass the absolute helper path, state path, identity tuple, a unique teammate dispatch identity, and the verbatim `phaseSkillLoad` manifest. The child must record its loads and pass `check-agent-write` with that unique identity before writing `research.md`.
16. Validate `research.md`, merge `phase: "research"` and `awaitingApproval: true` in interactive mode or `false` in exact quick mode, update progress, and present artifact approval when interactive.
17. In exact quick mode, record the quick bypass, generate missing artifacts in order, skip normal approval pauses, and continue into implementation in the same run.

## Prototype Reconciliation and Resume

After resolving the classified target and before normal resume or quick routing. Skip existing-current-spec recovery for new-spec intent:

1. Use `"$RALPH_CODEX_PLUGIN_ROOT/scripts/resolve_spec_paths.py"` and only its resolved `basePath`. When both `basePath` and `<basePath>/.ralph-state.json` exist, run `"$RALPH_CODEX_PLUGIN_ROOT/scripts/prototype_records.py" reconcile --base-path "$BASE_PATH" --state "$BASE_PATH/.ralph-state.json"`, then re-read state.
2. Treat a missing `activePrototypes` field as an empty map. Sort entries by `created`, then ID.
3. Resume an explicit active ID through `$ralph-specum-prototype --resume <id>` and stop this skill. In normal mode, resume the sole active entry automatically. When several remain, list deterministic IDs with question, status, blocker, `returnPhase`, and `returnTaskIndex`, then stop for an explicit ID.
4. In quick mode, ask no question. Sort entries that block design by `crea

Excerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

1. 2026-10-18
   !CAUTION B 89 / 100

   first audit · source a73b124937d9

Questions about Ralph Specum Start

• github.com/tzachbon/smart-ralph/tree/main/plugins/ralph-specum-c — the audited source at commit a73b124937d9 and every finding aboveObserved · 18 Oct 2026

Spotted something wrong? Report a correction · Methodology