docs: record that VS Code now auto-loads bundled plugin components

[devantler]

🤖 Generated by the Daily AI Engineer

Why

Our docs told VS Code users to do work they no longer need to do. The repo was built on the premise that VS Code is not a plugin surface, so anyone installing a plugin there was instructed to hand-copy agents into their workspace and hand-write MCP config. VS Code has since shipped agent plugins and now loads bundled skills, agents, and MCP servers automatically — so those instructions sent readers down a manual path for no reason. The README was also contradicting itself: it already documented a VS Code install path while the sections below it said VS Code could not install plugins.

What

Corrects the consumer-facing READMEs to say all three supported tools load bundled components automatically, and keeps the hand-written config only where it still applies — setups that are not installing a plugin at all. The two ADRs get dated superseding notes rather than edits-in-place, so the original point-in-time analysis stays readable. Two VS-Code-only caveats are now recorded in its place: the feature is Preview, and an organization can switch it off.

No manifest, skill, or workflow changes — our existing layout already matches what VS Code reads.

Fixes #66

[devantler]

🤖 Generated by the Daily AI Engineer

@codex review for factual accuracy of the VS Code capability claims against the cited official docs, and for any residual doc/ADR surface still asserting a manual VS Code delivery step.

[devantler]

🤖 Generated by the Daily AI Engineer

Readiness record

Verification (issue step 1 — "verify the full extent of VS Code agent-plugin support"): every capability claim in this PR was read verbatim off the live VS Code agent-plugins page, then re-fetched a second time to confirm the wording rather than trusting a search summary.

• Skills — skills field, "Defaults to skills/" ✅
• Agents — agents field, "Defaults to agents/" ✅
• MCP — "Place MCP server definitions in .mcp.json at the plugin root. VS Code discovers this file automatically when it loads the plugin." plus "start automatically when the plugin is enabled" and "implicitly trusted when you install the plugin … do not show a separate trust prompt at startup" ✅
• Caveats kept honest: chat.plugins.enabled is "managed at the organization level", and the feature is marked Preview. VS Code documents no minimum version, so none is claimed. mcpServers has no documented default (unlike skills/agents) — the auto-discovery claim rests on the quoted .mcp.json sentence, not on an inferred default.

Tried and evaluated as a user: the load-bearing question for this repo is whether our published layout actually resolves under those defaults — if any plugin declared skills/agents/mcpServers, the default would be overridden and the new docs would be wrong. Checked all 8 plugins by top-level manifest key: zero declare any of the three, and the directories/files sit exactly where the defaults point (skills/ in 8, agents/ in 3, .mcp.json at root in gitops-kubernetes). A negative control (injecting "skills": "custom/" into a copied manifest) flips the check to FAIL, so it has teeth.

Caveat stated plainly: I did not install the plugin in a live VS Code instance — that needs an interactive, org-gated Copilot session this unattended run cannot obtain. The claims are therefore grounded in the official contract plus a verified layout match, not an end-to-end install.

Validation: ./scripts/validate-manifests.sh → exit 0, 54 checks, 0 failures. No manifest, skill, or workflow changes. All CI green incl. CI - Required Checks; mergeStateStatus: CLEAN @ 589ec01.

Scope note: the README self-contradicted itself before this PR — it documented a VS Code marketplace install path while the sections below told readers VS Code could not install plugins. Both now agree.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 589ec01d1b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devantler]

🤖 Generated by the Daily AI Engineer

Claiming the 2 open Codex threads (scheduled tick ~14:2xZ) — verifying the VS Code agent-file discovery claim against both consumers' loaders, then fixing both.

[devantler]

🤖 Generated by the Daily AI Engineer

Both findings addressed at 21e63f0 (details on the threads). Validator self-test 44/44, shellcheck clean. Re-requesting review at the new head.

@codex review

[devantler]

🤖 Generated by the Daily AI Engineer

Releasing my 14:2x claim — 21e63f0 (pushed while I was building a docs-side fix) already addresses both threads with the stronger behavioural fix, and I am standing down. My uncommitted alternative (qualify the README/ADR claims instead of renaming) is discarded; the rename is the better resolution.

Independent verification of 21e63f0, from the differential:

• All 4 bundled agents renamed (automated-ai-engineer, portfolio-surveyor, flux-troubleshooter, vibe-coding-companion) — no straggler left on the old form.
• scripts/validate-manifests.sh rc=0 and its self-test rc=0 at that commit, run locally.
• No stale relative links to the old agents/<name>.md filenames anywhere in *.md/*.json (grep sweep).
• The README's Claude Code claim (filename-agnostic, name from frontmatter) matches Claude Code's sub-agents docs ("the filename doesn't have to match" — name comes from frontmatter name:).

One caution for the readiness gate: the cross-tool discovery claims are currently doc-inferred, not exercised — nobody has yet installed the renamed plugin in real VS Code, Claude Code, or Copilot CLI. I filed #74 for exactly that hands-on verification before this PR's claim ships as fact; linking it from the PR body would keep the trail. The tried-and-evaluated-as-a-user condition for this PR should include at least the Claude Code install check (cheapest of the three).

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 21e63f0cd6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devantler]

🤖 Generated by the Daily AI Engineer

Round-2 findings addressed at 04d1076 (details on the threads; self-test 44/44 with the new RED fixture). Re-requesting review at the new head.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Bravo.

Reviewed commit: 04d1076324

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devantler]

🤖 Generated by the Daily AI Engineer

Partial user evaluation @ 04d1076 — STAYING A DRAFT

Codex is green at head and the validator/self-test are exercised for real (44/44, RED-proven fixtures). But exercising Claude Code's own loader surfaced an unresolved naming risk: claude --plugin-dir …/gitops-kubernetes plugin details lists the renamed agent's component as flux-troubleshooter.agent — the filename stem, not the frontmatter name: — even though the docs say the frontmatter name is the identifier. If the runtime Agent-tool registry keys the same way, this rename silently changes every agent's invocation id (gitops-kubernetes:flux-troubleshooter → …:flux-troubleshooter.agent), breaking consumers — including the automated-ai-engineer plugin's portfolio-surveyor, which the run-loop skill invokes by exact name.

Blocked on: one decisive probe that needs an authenticated interactive session (this unattended shell cannot log in a nested claude -p): launch claude --plugin-dir <this-branch>/plugins/gitops-kubernetes and check whether the Agent tool exposes flux-troubleshooter or flux-troubleshooter.agent. If the stem wins, the fix is likely renaming to keep ids stable or teaching consumers the new id — do not promote before this is settled.

[devantler]

🤖 Generated by the Daily AI Engineer

Resolved the DIRTY state at d504e34 — merged origin/main (no force-push). Sole conflict was README.md: kept this branch's corrected premise (all three tools auto-load bundled components) while adopting main's 28-skill count and the new agent-improver agent; renamed the incoming agents/agent-improver.md to agent-improver.agent.md to satisfy this branch's suffix enforcement, and gave it the .agent.md README link like its siblings. validate-manifests.sh green, self-test 44/44 on the merged tree.

Re-requesting review at the new head.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d504e3488d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devantler]

🤖 Generated by the Daily AI Engineer

Claiming the two P2 findings from the 18:38Z Codex review @ d504e34 — fixing now (youthful-shirley, 19:1xZ).

[devantler]

🤖 Generated by the Daily AI Engineer

Round-3 P2s fixed at e50a2c0 (details on the threads). Re-requesting review at the new head.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e50a2c08e9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devantler]

🤖 Generated by the Daily AI Engineer

Round-4 P2s fixed at 5e5ee3a (details on the threads). Re-requesting review at the new head.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep it up!

Reviewed commit: 5e5ee3afc8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devantler]

🤖 Generated by the Daily AI Engineer

Status @ 5e5ee3a: pentad fully clear — promotion parked on ONE named blocker. Codex green at head (19:31Z, round 5), 0 unresolved threads, checks green, no conflict, premerge not-posted (Codex lane). The sole remaining readiness gap is the user-evaluation item recorded 14:29Z: hands-on confirmation on the three consumer surfaces (now scoped in #74 — agent-tool ID resolution for .agent.md files, YAML-list tools parsing, and per-surface built-in tool identifiers). That probe needs an interactive authenticated session — this scheduled run reconfirmed a nested claude -p is deauthed under cron, so an unattended run cannot clear it. Next interactive session: run the #74 verification, then promote this draft on the standard three conditions.