agentleFS
Sapere-IASapere-IA/sdd-onboarding-kit/skills/sdd-update/SKILL.md

sdd-update

Skill19 starsChanged 38 days agoMITView on GitHub

Update this project's SDD harness from the sdd-onboarding-kit. Use when the developer asks to update SDD, sync the harness with the kit, or apply a new kit version.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 5 are distinctive, and 5 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.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every migration step and every adapted-file merge requires developer approval before it is applied.

    line 15
  2. Distinctive:

    Never delete or overwrite a locally edited file without showing the developer what would be lost.

    line 16
  3. Distinctive:

    Treat the installed version as unknown: apply every changelog entry's migration whose changes are not already present (verify, don't …

    line 31
  4. Very specific:

    If the developer keeps a local checkout, use that path instead (read-only).

    line 39
  5. Very specific:

    If the installed file's current SHA-256 still equals the manifest's installed_sha256: overwrite with the new kit version (mechanical.

    line 50
  6. Distinctive:

    If not, the developer edited it locally: show the three-way situation (manifest version → local edits → new kit version) and ask.

    line 51
  7. Very specific:

    Gitignore: Add specs//feedback.md next to the existing SDD rendered-artifact entries in .gitignore.

    line 72
  8. Very specific:

    No mandatory step. If autonomy-policy.md is vendored under <harness-dir>/reference/, refresh it.

    line 79
  9. Very specific:

    Run scripts/validate-sdd-structure.sh (or .ps1) if present, with SDD_HARNESS_DIR=<harness-dir> when it is not .claude.

    line 88
  10. Distinctive:

    Delete the temporary clone.

    line 89

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SDD harness update skill

Bring this project's installed SDD harness up to date with the latest published version of the sdd-onboarding-kit, preserving every project-specific adaptation.

Kit source: {{KIT_REPO_URL}}

Safety rules (read first)

  • Touch only harness files: AGENTS.md (SDD sections only) and the CLAUDE.md stub, every harness directory listed in the manifest's harness block (.claude/, .codex/, .cursor/, .opencode/, .agents/), scripts/ harness scripts, spec templates, hooks installed by onboarding, and specs/ only for format migrations. Never touch product code.
  • Every migration step and every adapted-file merge requires developer approval before it is applied. Mechanical refreshes of unmodified verbatim assets may be batched under a single approval.
  • Never delete or overwrite a locally edited file without showing the developer what would be lost.
  • Work on a clean git state: if the repo has uncommitted changes to harness files, ask the developer to commit or stash first, so the update is a reviewable diff.
  • Clean up the temporary kit clone when done.

Procedure

1. Read the local manifest

Read <harness-dir>/sdd-kit-manifest.json (look in .claude/, .codex/, .cursor/, .opencode/, .agents/). It records kit_version, the harness block (primary harness and directory, instruction files, additional harness directories), and, per installed file, its kit source path, hashes, and adapted flag. A manifest without a harness block is a pre-2.1.0 Claude Code install (.claude/, CLAUDE.md).

If the manifest is missing (install predates the manifest), enter bootstrap mode:

  1. Fetch the kit (step 2) first.
  2. Detect the harness directories present, then reconstruct the file list by matching installed harness files against the kit's reference layout (<harness-dir>/agents/*, <harness-dir>/skills/sdd-workflow/*, templates, scripts, vendored <harness-dir>/reference/*). Files in a non-Claude agent format (Codex .toml, adapted frontmatter) are adapted: true.
  3. A file byte-identical to some kit version's source is verbatim; anything else is adapted: true.
  4. Treat the installed version as unknown: apply every changelog entry's migration whose changes are not already present (verify, don't assume), and flag uncertainty to the developer instead of guessing.

2. Fetch the kit

git clone --depth 1 {{KIT_REPO_URL}} <tmp-dir>

If the developer keeps a local checkout, use that path instead (read-only). Read the kit's VERSION and CHANGELOG.md.

3. Compare versions

  • Installed kit_version == kit VERSION: report "already up to date" and stop.
  • Otherwise, read every CHANGELOG.md entry newer than the installed version — these define both what changed and the migration steps.

4. Refresh verbatim assets

For each manifest record with adapted: false whose kit source changed:

  • If the installed file's current SHA-256 still equals the manifest's installed_sha256: overwrite with the new kit version (mechanical; batch these under one approval).
  • If not, the developer edited it locally: show the three-way situation (manifest version → local edits → new kit version) and ask.

5. Merge adapted files

For each adapted: true file whose kit source changed:

  1. Diff the kit source between the installed and new versions to isolate what the kit actually changed.
  2. Apply only those changes to the installed file, preserving the project adaptations (commands, paths, policies recorded in decisions/answers.md) and the harness packaging (agent frontmatter or Codex TOML, <harness-dir> paths). Apply the same change to every copy listed for additional harnesses.
  3. Show the resulting diff and get approval per file (related files may be grouped).

6. Run migrations

Execute the changelog Migration steps for each version being crossed, in order, each with approval. Never run a destructive step (deleting files, git rm) without showing exactly what it removes.

2.x → 3.0.0

The CHANGELOG.md entry stays authoritative; these are the steps it implies. Each step needs approval; apply every harness-directory step to each directory in the manifest's harness block.

  1. Renderer swap. Show and remove scripts/render-spec.mjs and/or scripts/render_spec.py (and their manifest records). Install scripts/render-spec.sh and scripts/render-spec.ps1 verbatim. Drop any renderer-runtime answer from decisions/answers.md (note it as superseded rather than deleting history).
  2. Spec assets and workflow. Refresh <harness-dir>/skills/sdd-workflow/templates/ (spec-shell.html.template, spec.css, spec.js and the trimmed *.md.template files), the sdd-workflow skill files and the agents from the kit (step 4 rules apply if they were edited locally). Existing specs keep rendering unchanged; they are not rewritten to the new templates.
  3. Stale rendered pages. List the per-document .html files under specs/ (one per markdown doc, from the 2.x renderer), delete them after approval, then re-render each spec folder with sh scripts/render-spec.sh specs/<feature-slug>/ so each spec has a single spec.html. If the project commits rendered HTML (decisions/answers.md), tell the developer the deletions and new pages need committing; do not commit them yourself.
  4. Gitignore. Add specs/**/feedback.md next to the existing SDD rendered-artifact entries in .gitignore.
  5. Session skills. Install skills/bro/SKILL.md → <harness-dir>/skills/bro/SKILL.md and skills/closing/SKILL.md → <harness-dir>/skills/closing/SKILL.md (new verbatim records in the manifest).
  6. AGENTS.md. In the spec-storage section, replace the old render command with sh scripts/render-spec.sh specs/<feature-slug>/ (one page specs/<feature-slug>/spec.html) and add the one-line feedback.md flow; add the "Session skills" section (bro, closing) from the kit template. Merge as an adapted file (step 5 rules).
  7. Validate. In step 7 below, the structure validator now also requires scripts/render-spec.sh and the bro and closing skills.

3.0.0 → 3.1.0

No mandatory step. If autonomy-policy.md is vendored under <harness-dir>/reference/, refresh it. Install the goblin-mode pack only if the developer asks for it (CHANGELOG.md 3.1.0 Migration step 2).

3.1.0 → 3.2.0

Refresh the closing skill in every harness directory, the vendored memory-policy.md and, if installed, goblin-mode; update the closing line in AGENTS.md (CHANGELOG.md 3.2.0 Migration).

7. Rewrite the manifest and report

  • Rewrite <harness-dir>/sdd-kit-manifest.json: new kit_version, updated_at, the harness block, fresh hashes for every touched file; add records for newly installed files.
  • Run scripts/validate-sdd-structure.sh (or .ps1) if present, with SDD_HARNESS_DIR=<harness-dir> when it is not .claude.
  • Delete the temporary clone.
  • Report: previous → new version, files refreshed mechanically, files merged, files skipped (locally edited, developer declined), migration steps run, and anything left for the developer.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.