Historical roadmap: maintained skills before plugin consolidation

🤖 Generated by the Agentic Engineer

Product and current baseline

devantler-tech/agent-skills is the tool-neutral catalogue of reusable skills consumed individually by supported assistants and through Agent Plugins. The README is the maintained catalogue and installation entry point.

Verified on 2026-10-03 at main 2e0a095729fa49a100dfb382434b548fdb4457f8, published as v1.21.15: 28 indexed skills — 11 maintained here and 17 upstream pointers. The maintained skills cover agentic engineering, repository instructions, release and engineering practices, tenant onboarding, and conversational product work. The current index consistency check passes. A fresh native tag-pinned installation independently matches all 11 authored skill bodies, 22 supporting files and repository/path/tree/pin provenance; the installer serializes frontmatter and trims blank lines at the body boundary, while the remaining body content and every resource match the release Git objects.

Installation uses gh skill install or the batch helper at scripts/install.sh; the installation guide documents supported agents and options. This authored-skill installation does not establish the health of every external catalogue target or every runtime marketplace discovery path.

Delivery themes

Reliable installation and discovery. README index validation checks local targets and agreement between each row's source, name, and install command. The scheduled upstream check verifies external skill targets. Keep each defect grounded in an observable install or discovery failure; distinguish transport outages from malformed or missing targets.
Publication bound to validated content. Release correctness binds the expected repository, checkout commit, clean files, validation result and remote published tag. The release gate and its hermetic checks own this contract; independent tag and consumer observations establish the published outcome. A successful command exit alone is insufficient.
Curated reusable skills. The contribution and inclusion criteria are the quality bar. Prefer an existing maintained upstream where it fits. Extract a new generic skill only after demonstrated use in at least two products establishes a reusable, tool-neutral contract. Catalogue growth follows demonstrated need rather than a numerical target.
Evidence and human accountability for unfamiliar choices. Engineering and self-improvement procedures carry reusable evidence evaluation and layered accountability briefs. Structural checks never grant authority or replace semantic review, current-head readiness or actual human comprehension. The format and integrations for Make unfamiliar engineering decisions understandable to accountable humans #620 are delivered; its remaining maintainer pilot requires a real participant answering the six operational questions from an actual revision-bound brief, with consent before recording misunderstandings. Keep that criterion open until observed.

Ownership and evidence

This repository owns its authored skills and the catalogue. agent-plugins owns bundle composition, plugin-authored agents, manifests and installation adapters; its updater synchronizes maintained skill sources with recorded provenance, and CI checks bundle ownership, manifest parity, versions and desired-state digests. Generic skill fixes start in their owning source rather than in a generated bundle or runtime cache.

Native child issues track actionable work, with implementing PRs linked to those children. Completed work and validation evidence remain in linked issues and comments; this roadmap describes current product direction rather than duplicating a status board. Consumer plugin pins and runtime installation currency are verified in their own deployment lifecycle.

🤖 Generated by the Daily AI Assistant

Decomposed this roadmap into independently-shippable children:

devantler-tech/agent-skills#34 — Theme 1: grow the in-house skill set from a proven portfolio pattern (feat, per skill)
devantler-tech/agent-skills#35 — Theme 2: validate that each upstream install target resolves (ci; the existing check-readme-index.sh already guards the in-house subset)
devantler-tech/agent-skills#36 — Theme 3: codify the skill inclusion / quality-bar criteria (docs) → implemented in devantler-tech/agent-skills#37 (draft)

Theme 3 picked up first (lowest-risk, fully additive); devantler-tech/agent-skills#37 awaits promotion.

🤖 Generated by the Daily AI Assistant

Strategy review devantler-tech/agent-skills#2 — `agent-skills` (2026-06-24)

Second strategy review of this roadmap. Verified against live main and the GitHub API; the epic body is refreshed to match (see the 2026-06-24 footer note).

Where the product is now (vs. when this epic was filed, 2026-06-08)

	At filing (06-08)	Now (06-24)
Indexed skills	19	22
In-house (authored)	1 (`ways-of-working`)	4 — `agent-instructions`, `conventional-release`, `gitops-tenant-onboarding`, `ways-of-working`
Open children	3	0 (all closed/merged)

The index lockstep CI (PR-blocking) + the scheduled upstream-resolution check (#38) are both live, so the catalogue cannot silently drift between maintenance runs.

Theme outcomes

Theme 1 — grow the in-house skill set: PHASE 1 DELIVERED, reframed to opportunistic. Child devantler-tech/agent-skills#34 closed; the three obvious extra patterns landed since filing — gitops-tenant-onboarding (feat: bundle the first custom agent (agents/) into a plugin — Theme 1 #45), agent-instructions (feat(plugins): add discovery keywords to each plugin.json (Theme 3 first child) #47), conventional-release. The remaining devantler-tech patterns either already live inside one of the four in-house skills, or are too product-specific / not-yet-generalisable to clear the devantler-tech/agent-skills#37 inclusion bar. New direction: add a new in-house skill only when a fresh generic pattern proves out across 2+ products (the holistic shared-library trigger) — evidence-gated, not scheduled. Curate quality over count; a forced fifth skill without a proven, non-overlapping pattern would fail the devantler-tech/agent-skills#37 "generic & reusable" bar.
Theme 2 — harden the index against install-target drift: DONE. Closed on all three fronts — in-house on-disk resolution (docs: reconcile AGENTS.md 'single source of skills' with the live multi-upstream sourcing (and bundle agent-skills' own skills?) #33), cross-column consistency (ci: guard that every bundled SKILL.md carries upstream provenance #41), and scheduled upstream resolution (Roadmap: agent-plugins marketplace — reach & feature-completeness #38). Steady-state.
Theme 3 — codify the inclusion / quality bar: DONE. devantler-tech/agent-skills#36 → devantler-tech/agent-skills#37 merged 2026-06-10; the criteria doc is referenced from AGENTS.md.

Verdict

agent-skills is in a healthy steady-state: governed, agent-neutral, fully CI-guarded, with the curated in-house value built out. There is no forced advance work here right now — and per the contract that is a legitimate outcome for a mature shared library; manufacturing a fifth in-house skill or a redundant check would be filler, not value.

This epic stays open as the standing strategy anchor. Re-pickup triggers (so the next run knows when there is work):

A new generic pattern proves out across 2+ products → extract it as in-house skill devantler-tech/agent-skills#5 (Theme 1, evidence-gated).
A valuable upstream skill appears that fills a real portfolio gap → add an upstream-pointer row.
The scheduled upstream-resolution check hard-fails (a 404 on a pinned ref) → repin/replace that row.
The maintainer decides to automate skills↔plugins consistency (still deferred — plugins scope is settled via docs: record MCP-server-bundling decision (skills only) via ADR-0001 #15).

No code PR this tick: the right work for a healthy, fully-delivered roadmap is this reconciliation + direction refresh, not a forced change.

🤖 Generated by the Daily AI Engineer

Strategy review devantler-tech/agent-skills#3 — `agent-skills` (2026-07-17)

Third review of this roadmap, verified against live main and the GitHub API.

Where the product is now (vs. review devantler-tech/agent-skills#2, 2026-06-24)

	06-24	Now (07-17)
Indexed skills	22	28
In-house (authored)	4	10
Open issues	1 (this epic)	2 (this epic + roadmap devantler-tech/agent-skills#56)
Open PRs	0	0

What shipped since (all evidence-gated, matching Theme 1's refreshed bar)

ADR-0002 — the generic AI-engineer brain, extracted (feat: add the automated-ai-engineer plugin (agents+skills bundle per ADR-0002) #63 self-improvement, devantler-tech/agent-skills#66 product-engineering, devantler-tech/agent-skills#67 portfolio-maintenance): the portfolio's proven Daily-AI-Engineer skill set generalized into tool-neutral shared skills — the single highest-leverage pattern the library has absorbed, plus devantler-tech/agent-skills#68 routing skill improvements to each skill's canonical upstream.
ADR-0003 — the vibe-coding family (Fix stale AGENTS.md Validation step 3 (.skills=="skills/" contradicts auto-discovery convention) #65: needs-stack-mapping, allowed-stack-guardrail, jargon-free-voice), now anchored by its own roadmap epic devantler-tech/agent-skills#56 — that direction is tracked there, not here (no double-tracking).
Hardening: self-tests for the index-lockstep guard (Deliver reusable Agentic Engineer definitions through canonical sources #51), installer (chore(deps): update agent skills #52), and upstream guard (fix: omit skills/agents path fields so plugins install in Claude Code #53); Release-run serialization fixing a same-tag race (Verify Copilot CLI discovers bundled agents shipped as agents/*.md (vs .agent.md) #64); docs repointed off the archived reusable-workflows (docs(adr): design ADR for the Vibe Coding plugin (shape, skill family, stack guardrail, voice) #57/docs(adr): design the Vibe Coding plugin (shape, skill family, stack guardrail, voice) #58).

Theme status

All three original themes remain DONE; Theme 1's "opportunistic, evidence-gated" framing is working exactly as designed — both post-review additions arrived through ADRs proving cross-portfolio patterns, not schedule-driven authoring.

Direction (refreshed)

This epic stays the general-library anchor; the vibe-coding family direction lives in devantler-tech/agent-skills#56. Steady state here = curation quality over count — no forced new theme.
Distribution watch-item: the agent-plugins rescope (plugins#7) is the bundling/marketplace surface for these skills; index changes must stay bundle-compatible while that lands.
The previously maintainer-deferred skills↔plugins consistency item is visibly moving maintainer-side: an org App (botantler-1[bot]) has begun opening chore(deps): update agent skills PRs on consumer repos — evidence the sync-automation call was made. Its trust-gate standing is still an open maintainer decision tracked in the run reports; once settled, that deferred item can be closed out here.

No new children filed — nothing evidence-backed is startable right now, and the quality bar (#37) argues against manufacturing additions. Next review on the normal cadence.

🤖 Generated by the Daily AI Assistant

Current product review, 2026-09-06: the catalogue contains 28 skills, 11 maintained here. The existing index checks, full shell suite, publication dry run, and current main CI are green. The three vibe-coding skills are delivered; devantler-tech/agent-skills#56 is now closed with its implementation evidence.

The useful next direction is dependable installation and delivery of the current catalogue. Publication queue adoption shipped in devantler-tech/agent-skills#108. Hands-on installer assessment found that help and malformed arguments entered the bulk install path, with environment glob expansion and no installation-path tests; devantler-tech/agent-skills#109 records the defect and is now a real child of this strategy issue. Its delivery PR devantler-tech/agent-skills#110 adds offline help, validates before side effects, and exercises exact install arguments and partial failures.

Success proxy for this slice: every usage-error fixture makes zero gh calls; a valid two-skill/two-agent fixture makes four intended calls; partial failures preserve diagnostics and attempt remaining installs. These pass locally. Final current-head review and hosted CI remain part of delivery. Keep this issue as the standing strategy anchor and retain the evidence bar for adding new skills: proven reusable needs, not catalogue growth alone.

🤖 Generated by the Daily AI Assistant

Second-round delivery, 2026-09-06: devantler-tech/agent-skills#111 is closed by merged devantler-tech/agent-skills#112 at 9f8a01cf9b8d1a59c51b748b82f43380e6da06b2. Upstream checks now require a matching SKILL.md file response; malformed successful payloads cannot claim health or disappear into transport warnings. All 17 focused cases, the full shell suite, and main CI passed. The prior installer delivery devantler-tech/agent-skills#110 is also merged and green.

The next verified distribution gap is devantler-tech/agent-skills#113, now a real child of this roadmap and in Backlog. Offline fixtures show the fresh-publication path reaches gh skill publish without binding the checkout repository or commit to its expected inputs. No bad live release was observed. Establish the real CLI selection behavior before choosing that guard; existing publication idempotency remains covered.

🤖 Generated by the Agentic Engineer

Delivered the dependency-refresh slice end to end. agent-skills#148 merged and published as v1.20.1; agent-plugins#302 merged the concrete refresh repair; the native updater generated agent-plugins#304, which merged as a41b9b6 with agentic-engineering version 5.6.6.

Independent native installations of both published source and merged bundle matched their Git objects: 11 authored skills/19 support files and 26 bundled skills/98 support files. The corrected guidance matches the source tree 7c3ee27056f5a9dac57b53565cdc2facc09cea53. Required current-head CI and substantive reviews passed before all three merges. The refresh repair passed 60 regression scenarios and an actual GitHub branch-update evaluation that retained existing commits and merge settings and observed new-head PR CI.

This completes devantler-tech/agent-skills#147 and #303. Published/installed artifact proof is separate from existing runtime adoption. The human-comprehension pilot, VS Code/Copilot integration evaluations and automatic release rollout retain their existing independent acceptance gates.

🤖 Generated by the Agentic Engineer (interactive Claude Code session)

Strategy review, 2026-10-17

Where the catalogue is. Release v1.26.0; 89 commits since 2026-08-01. Recent work added documentation skills (#276, devantler-tech/agent-skills#278, devantler-tech/agent-skills#282), the upstream release review (#280) and Kanban value pull (#272). All 81 children of this roadmap are closed, so it had no open work left to pull. The early-October run of about 40 fixes to the installer, catalogue and evidence scripts has stopped; there are no open PRs.

What the review found.

The three skills that every engineer and improver run loads are the largest: agent-improvement 41 KB, portfolio-maintenance 34 KB, product-engineering 23 KB. All three are over the Agent Skills guidance of about 5,000 tokens, and nothing stops them growing. This is the clearest cost and reliability lever in the repository.
There is no record of anyone outside the portfolio using the catalogue, although the README is written for them. Theme 3 ("growth follows demonstrated need") cannot be applied without that signal.
The upstream release review for the 17 upstream catalogue entries runs in agent-plugins, where those skills are bundled and pinned; see the review on Roadmap: agent-plugins marketplace — reach & feature-completeness #38.

Roadmap, in priority order:

devantler-tech/agent-skills#283: keep authored skills within the context budget. First devantler-tech/agent-skills#285 (a CI size ratchet; refined, no prerequisites), then devantler-tech/agent-skills#286 (split agent-improvement, waits on devantler-tech/agent-skills#285).
devantler-tech/agent-skills#284: measure outside adoption, two hours, then decide whether outside reach is worth investing in.
Make unfamiliar engineering decisions understandable to accountable humans #620: only the maintainer comprehension pilot remains.

Decision for the maintainer: #620 can only close if you take part. Either run the pilot (read one real accountability brief and answer its six operational questions, about 15 minutes), or tell us to close it with the pilot recorded as not done.

🤖 Generated by the Agentic Engineer (interactive Claude Code session)

Outside adoption measured: internal-first for now (#284)

The two-hour measurement in devantler-tech/agent-skills#284 found almost no outside use: 8 unique visitors in 14 days, no external referrers, two outside stars and one install of one skill on a public skill directory. The portfolio's own runs and the agent-plugins bundle are the only demonstrated consumers.

So the catalogue is treated as an internal-first library. It stays public, spec-conformant and agent-neutral, but outside reach (directory listings, installation docs for outsiders, catalogue growth for outsiders) is not invested in until demand shows up. The trigger to revisit, and the baseline to compare against, are on #284. Re-read those signals at the next strategy review.

Roadmap order is otherwise unchanged: devantler-tech/agent-skills#283 (size budget, with devantler-tech/agent-skills#285 in review as devantler-tech/agent-skills#293) stays first.

🔗 Related PRs

#590 - ci: scope engineering-practice skill updates [closed]
#597 - fix(ci): preserve safe skill pins across update scopes [merged]
#613 - chore: adopt the Marketplace skill updater [merged]

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

Create Plan

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

🤖 Generated by the Agentic Engineer

The maintainer requested consolidation into Agent Plugins in devantler-tech/monorepo#4742. This historical roadmap and its completed child inventory were transferred intact from agent-skills. The maintained product roadmap is #38.

The real maintainer comprehension pilot remains open as #620 with its original acceptance criteria and discussions. Closing this superseded product roadmap does not accept or close that pilot.

Navigation Menu

Repository navigation

Historical roadmap: maintained skills before plugin consolidation #621

Product and current baseline

Delivery themes

Ownership and evidence

Strategy review devantler-tech/agent-skills#2 — `agent-skills` (2026-06-24)

Where the product is now (vs. when this epic was filed, 2026-06-08)

Theme outcomes

Verdict

Strategy review devantler-tech/agent-skills#3 — `agent-skills` (2026-07-17)

Where the product is now (vs. review devantler-tech/agent-skills#2, 2026-06-24)

What shipped since (all evidence-gated, matching Theme 1's refreshed bar)

Theme status

Direction (refreshed)

Strategy review, 2026-10-17

Outside adoption measured: internal-first for now (#284)

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Historical roadmap: maintained skills before plugin consolidation #621

Description

Product and current baseline

Delivery themes

Ownership and evidence

Activity

devantler commented on Jun 10, 2026

devantler commented on Jun 24, 2026

Strategy review devantler-tech/agent-skills#2 — agent-skills (2026-06-24)

Where the product is now (vs. when this epic was filed, 2026-06-08)

Theme outcomes

Verdict

devantler commented on Jul 17, 2026

Strategy review devantler-tech/agent-skills#3 — agent-skills (2026-07-17)

Where the product is now (vs. review devantler-tech/agent-skills#2, 2026-06-24)

What shipped since (all evidence-gated, matching Theme 1's refreshed bar)

Theme status

Direction (refreshed)

devantler commented on Sep 6, 2026

devantler commented on Sep 6, 2026

devantler commented on Oct 1, 2026

devantler commented on Oct 17, 2026

Strategy review, 2026-10-17

devantler commented on Oct 24, 2026

Outside adoption measured: internal-first for now (#284)

coderabbitai commented on Nov 1, 2026

devantler commented on Nov 1, 2026

Metadata

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Issue actions

Strategy review devantler-tech/agent-skills#2 — `agent-skills` (2026-06-24)

Strategy review devantler-tech/agent-skills#3 — `agent-skills` (2026-07-17)