agentleFS
gleicongleicon/devskills/skills/ds-retro/SKILL.md

ds-retro

Skill20 starsChanged 46 days agoMITView on GitHub

Post-release retrospective — compare what SPEC.md/GRILL.md decided against what the release range actually shipped.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    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
  1. Very specific:

    When invoked, run a post-release retrospective over a release range: compare what was decided (SPEC.md, GRILL.md) against what shipped, and …

    line 7
  2. Distinctive:

    no tags: on a non-default branch, offer merge-base-with-default-branch..HEAD (a branch retro).

    line 17
  3. Distinctive:

    Read whatever exists; open the report with a source inventory naming what was found and what analysis its absence disables ("GRILL.md ✓ · …

    line 22
  4. Very specific:

    Pinned reads: read SPEC.md/GRILL.md at the range end — git --no-pager show vNEXT:SPEC.md — never the working tree.

    line 25
  5. Very specific:

    Check each commit is a dedicated docs(spec): / docs(grill): commit and the commit count squares with the amendment lines found in the files.

    line 33

≈ “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

When invoked, run a post-release retrospective over a release range: compare what was decided (SPEC.md, GRILL.md) against what shipped, and distill rules for the next cycle. The loop-closer to /ds-spec and /ds-grill-me — they record decisions going in; this judges them coming out.

Arguments

  • A release range, resolved in order:
    • vPREV..vNEXT — used verbatim; any valid git range works, branches and SHAs included (v0.4.0..HEAD for a retro before tagging).
    • a single ref — the range is previous-version-tag..ref.
    • nothing — inspect the repo and offer a range; never silently pick one:
      • HEAD ahead of the newest version tag (git --no-pager tag --sort=-v:refname) → offer newest-tag..HEAD — the unreleased work, the pre-merge case — with the last two tags as the named alternative.
      • HEAD at the newest tag → the last two version tags.
      • no tags: on a non-default branch, offer merge-base-with-default-branch..HEAD (a branch retro); on the default branch, ask for an explicit range. Never guess.
  • --record [path] — also write the retro to RETRO.md in the current directory, or to path. Without the flag, the retro lives in the session only.

Sources

Read whatever exists; open the report with a source inventory naming what was found and what analysis its absence disables ("GRILL.md ✓ · SPEC.md ✗ — spec-break analysis skipped").

  • Floor: at least one of SPEC.md / GRILL.md. Neither → stop and point at /ds-spec and /ds-grill-me; with no recorded decisions there is nothing to compare, and the output would be a changelog, not a retro.
  • Pinned reads: read SPEC.md/GRILL.md at the range end — git --no-pager show vNEXT:SPEC.md — never the working tree; the next cycle may already be amending them. Mid-implementation amendments are the in-file dated Amended YYYY-MM-DD: lines whose dates fall inside the range (tag dates come from git --no-pager log --no-color -1 --format=%ci <tag>).
  • Enrichment: .project/roadmap.md and .project/state.md (# settled / # hazards) from the working tree — git-ignored, so the working copy is the only copy. Flag them in the report as "current lens, not range-pinned."

Discipline audit

The amendment discipline (scaffolded by devskills init --spec-discipline) makes the decision files self-describing; git's job is to verify the files, not to be the source:

  1. git --no-pager log --no-color --format='%h %s' vPREV..vNEXT -- SPEC.md GRILL.md.
  2. Check each commit is a dedicated docs(spec): / docs(grill): commit and the commit count squares with the amendment lines found in the files.
  3. A mismatch — spec edits riding feature commits, changes with no Amended: line — is itself a retro finding: report the drift. Only then read each offending commit with git --no-pager show --no-ext-diff --no-color --format='%h %s' <sha> -- SPEC.md GRILL.md to recover what changed.

Report

Compare decided vs happened across:

  • Assumptions measurement killed — assumptions the release's reality disproved.
  • Decisions that held — calls that survived contact, and why.
  • Spec breaks — shipped behavior that contradicts an unamended decision.
  • Accepted-consequence accuracy — did the costs knowingly accepted show up as predicted?
  • Unvalidated-at-scale leftovers — decisions still untested by real use, carried into the next cycle.

Plus the source inventory and any discipline-drift findings.

RETRO.md (with --record)

  • File shape: a # Retrospectives header with one intro line; one section per retro, newest first, inserted directly under the header.
  • Section shape: ## vPREV..vNEXT — YYYY-MM-DD, distilled rule bullets first, then ### Evidence with condensed support. Rules-first is what lets /ds-grill-me and /ds-spec read the top of a section and stop.
  • Idempotent per range: re-running for a range already present replaces that range's own section; other sections are never touched — git history is RETRO.md's audit trail.

Companion habit

/ds-grill-me and /ds-spec read RETRO.md at invocation (current directory, then repo root) and surface its rules as priors — cited when they bear on the question at hand, overridable by the user, never silently binding.

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.