spec-to-issue
Skill1 starsChanged 7 months agoMITView on GitHub
Set up loop-runnable work from a spec: put the spec where the executor can ALWAYS reach it (the GitHub issue body), optionally open a draft review-PR that is closed-not-merged when work begins, and instruct the executor to re-materialize the spec onto its branch if durable. Feeds full-path-github. Use when the user says "turn this into an issue", "set up this spec for the loop", "groom this for the orchestrator", or hands you a spec/plan to delegate.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- anyoneanderson/agent-skillstheir own version, first seen Oct 13, 2026
- buemura/claude-code-skillstheir own version, first seen Sep 9, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- anyoneanderson/agent-skills★ 3same name
- buemura/claude-code-skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository18 other files it gives its agents.
- close-out-sessionSkill
- code-reviewSkill
- comprehension-artifactSkill
- continue-workflowSkill
- enforce-integration-testSkill
- extract-engineering-styleSkill
- full-path-githubSkill
- generate-skill-docSkill
- gh-pr-createSkill
- grill-meSkill
- improve-codebase-architectureSkill
- playwright-capture-flowSkill
- prd-to-issuesSkill
- scaffold-docs-folderSkill
- scaffold-workspaceSkill
- setup-nvim-tmuxSkill
- structured-backlogSkill
- tdd-pocockSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 50
When a launch prompt (or a spec's instructions) needs a shared skill, include this preamble so the session can always reach it:
- Distinctive: line 60
Only clone what the work actually needs.
- Distinctive: line 105
Before filing, for EVERY path/repo the spec references, confirm the executor can reach it from the repo it will be scoped to:
- Very specific: line 127
Only if the user wants a review pass before work starts.
- Distinctive: line 132
If you must touch a local repo (e.g.
- Distinctive: line 133
NEVER work in a dirty checkout or on someone's feature branch:
- Very specific: line 134
Check git -C <repo> status --short and git branch --show-current.
- Very specific: line 135
If dirty / on a feature branch → create an isolated worktree off origin/main (git worktree add <tmp> -b <branch> origin/main) and work …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Spec → Issue (work setup that feeds full-path-github)
This is the pre-intake step. full-path-github starts from a groomed
issue; this skill produces that issue from a spec, in a way that avoids the
recurring discoverability dead-end (the executor told to fetch something it can't
reach) and the dirty-local-git trap.
The core rule — the issue body IS the spec
A cloud / headless session is guaranteed to have one thing: the issue it's pointed at. Anything else (a spec file in another repo, a vendored doc, a local-only clone) may be unreachable. So:
The full spec goes in the issue body. Not a link to a spec file. Not "see the spec PR." The text itself. Reachability becomes a non-problem because there is nothing to fetch.
Corollaries:
- Small reference code to port goes inline too — as a fenced code block in
the issue (GitHub blocks
.pyattachments and has no attachments API, so attachments don't help). A few hundred lines in a```pythonblock is fine and fully reachable. Large source belongs in the repo, not the issue. - Never reference a path in a local-only clone (no remote) or in a different repo than the one the executor is scoped to. Verify reachability (see Step 3).
Reaching SHARED skills (workshop) from a non-laptop instance
Reusable skills live in one source of truth — jwrobes/workshop (and
jwrobes/skills for the dev/orchestrator skills). How an instance reaches them
depends on the instance, and there are three types (not two):
| Instance | Skill access | Mechanism |
| Laptop (interactive) | ~/.claude/skills/ symlinks → the workshop/skills checkouts | already set up; add a symlink per new shared skill |
| Cloud / web (claude.ai/code) | repo-scoped, no shared FS | clone at runtime |
| Server cron (headless) | bare ~/.claude, NO skills dir, NO local checkout | clone at runtime (same as cloud) |
So clone-at-runtime is the portable mechanism for any non-laptop instance. When a launch prompt (or a spec's instructions) needs a shared skill, include this preamble so the session can always reach it:
# Shared skills live in their own repos, NOT in this work-repo. Clone what you need:
git clone --depth 1 https://github.com/jwrobes/skills.git /tmp/skills # dev/orchestrator skills
git clone --depth 1 https://github.com/jwrobes/workshop.git /tmp/workshop # shared workshop tools/skills
# then read e.g. /tmp/skills/full-path-github/SKILL.md or /tmp/workshop/<skill>/SKILL.mdOnly clone what the work actually needs. One source of truth, zero drift, zero per-repo vendoring — the cost is that the prompt must include the clone line, so make it boilerplate. (Vendoring a skill into the work-repo is the alternative — zero-setup reach but drift risk; reserve it for skills a session can't function without and that change rarely, like the orchestrator.)
The spec-PR is a review surface, not an artifact
A spec doesn't need a PR — but a PR is a far better review surface (inline comments, rendered diff, a thread). So the spec-PR is optional and review-only:
- Open it as a DRAFT, titled
SPEC (review only — do not merge). - Review/comment there; sync accepted edits back into the issue body (the source of truth).
- Close it — do NOT merge it — when implementation begins. Its job was review; merging it would commit a planning doc into the repo (clutter + reachability risk) and never-merging it would leave a zombie PR (the exact "unhealthy state" the fleet dashboard exists to kill).
- The spec's durable home is the impl branch (next rule), not this PR.
Re-materialize the spec onto the impl branch (if durable)
Tell the executor: if the spec is worth keeping with the code, write it into
the implementation branch as a docs/ or plans/ file as part of the work — so
the spec ships, reviewed, on the real (merged) PR, co-located with what it
describes. This dovetails with the comprehension-artifact gate. Trivial specs
need not be materialized; say so.
Steps
Step 1: Assess the spec
- Is it groomed enough to run? (what / where / how-to-verify — the vagueness
gate
full-path-githubwill apply anyway). Tighten if not. - Does it cite reference code to port? Decide: small → inline in the issue; large → it must live in the target repo (commit it there first), never in a local-only/other repo.
Step 2: Decompose if needed
Multi-part work → parent tracker + native sub-issues (POST issues/{n}/sub_issues
-F sub_issue_id=<int>), blocked_by deps (-F issue_id=<int>), ready on the
unblocked leaf. (Same conventions as full-path-github guardrails — integer
-F form; the string form silently no-ops.)
Step 3: Reachability pre-flight (the dead-end guard)
Before filing, for EVERY path/repo the spec references, confirm the executor can reach it from the repo it will be scoped to:
# Is a referenced repo a local-only clone (no remote)? → unreachable, inline or vendor it.
git -C <referenced-clone> remote -v # empty = local-only = NOT reachable
# Is the referenced file in a DIFFERENT repo than the executor's target? → unreachable.Anything unreachable → inline it into the issue (spec text, or small code in a code block) or commit it into the target repo first. This is the step that prevents the #105 / vendored-skill / v1-collector dead-ends.
Step 4: File the issue(s) with the spec inline
Issue body = the full spec + any inline reference code block + (for leaves) the acceptance criteria. Add the standing instruction:
Your spec is THIS issue body. Implement against the repo as it really is (ground every cited path). If this spec is worth keeping, write it into your branch under
docs/orplans/as part of the work.
Step 5: (Optional) Open the draft review-PR
Only if the user wants a review pass before work starts. Draft, titled SPEC
(review only — do not merge), body = the spec (or a pointer to the issue).
Record that it must be closed, not merged, when implementation begins.
Step 6: Handle dirty local git (don't disturb in-flight work)
If you must touch a local repo (e.g. to commit reference code into the target), NEVER work in a dirty checkout or on someone's feature branch:
- Check
git -C <repo> status --shortandgit branch --show-current. - If dirty / on a feature branch → create an isolated worktree off
origin/main(git worktree add <tmp> -b <branch> origin/main) and work there. Leave the existing checkout untouched. Push the new branch; open a PR.
Step 7: Hand off
Report the issue number(s) + the launch instruction for full-path-github
(point it at the ready leaf; it clones the orchestrator skill or reads the
vendored copy). Note any spec-PR that must be closed-not-merged at impl start.
Output
- Issue(s) whose body IS the spec, fully reachable (nothing to fetch).
- Native sub-issue tree + deps if decomposed;
readyon the unblocked leaf. - Optional draft review-PR, flagged close-not-merge.
- Reference code inlined or committed into the target repo (never local-only).
- A clean hand-off to
full-path-github.
Composition
Feeds full-path-github (this produces the groomed issue it consumes). Pairs
with structured-backlog (decomposition) and the comprehension-artifact gate
(the spec re-materialized on-branch is the same low-comprehension-debt instinct).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

