Skip to content

docs: agentic software factory inception (docs/agent-guidance-inception.md) - #4861

Open
cliffhall wants to merge 11 commits into
v2/mainfrom
v2/docs/4859-agent-guidance-inception
Open

cliffhall wants to merge 11 commits into
v2/mainfrom
v2/docs/4859-agent-guidance-inception

Conversation

@cliffhall

Copy link
Copy Markdown
Member

Closes #4859

Part of the agentic software factory tracker #4858.

What this adds

docs/agent-guidance-inception.md inventories the MCP Inspector's agentic software factory and decides how each part carries over to this repo. Sources: inspector#2498 and the Inspector's v2/main at 64a50d6f.

How this was tested

This is a docs-only change. Every repo fact was checked against live state with gh and git: labels, milestones, board #43 fields, repository and security settings, workflows, and per-server tooling.

🤖 Generated with Claude Code

Inventory the MCP Inspector's agentic software factory (AGENTS.md rules,
skills, gate scripts, CI and release workflows, board conventions), give each
element a transfer/adapt/N/A verdict for this repo, map every CLAUDE.md
section and #4473 decision to a destination, list reusable templates, and
propose the wave-ordered sub-issues of #4858.

Closes #4859

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall cliffhall added the v2 label Sep 27, 2026
This was referenced Sep 27, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🟡 Changes recommended

The plan contains contradictory acceptance criteria, an inaccurate protocol statement, and incomplete action-pinning coverage.

Review effort: Balanced
Findings: 1 High severity · 6 Medium severity · 1 Low severity

Open (8)
What changed in this PR

Documents how the Inspector’s agentic software factory could be adapted for this two-language server monorepo.

Changes:

  • Inventories rules, skills, gates, workflows, and project conventions.
  • Maps existing guidance and related issues into a phased migration plan.
  • Defines 13 follow-up workstreams across rules, CI, security, releases, and automation.
File Description
docs/​agent-guidance-inception.md Adds the factory assessment and implementation roadmap.

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

Comment thread docs/agent-guidance-inception.md Outdated
Comment thread docs/agent-guidance-inception.md Outdated
Comment thread docs/agent-guidance-inception.md Outdated
Comment thread docs/agent-guidance-inception.md Outdated
Comment thread docs/agent-guidance-inception.md Outdated
Comment thread docs/agent-guidance-inception.md Outdated
Comment thread docs/agent-guidance-inception.md Outdated
Comment thread docs/agent-guidance-inception.md Outdated
- Scope client evidence to server-facing changes, with a targeted probe
  otherwise, and require both clients consistently
- HTTP+SSE is deprecated, not removed, in the 2026-07-28 spec
- Scope labels only where an issue concerns one server
- S7 depends on S5 and S8
- Release prep is up to three PRs under #4472's two bump PRs
- Action pinning covers claude.yml too
- Sweep dry runs write nothing

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 1: 8 findings, all fixed in 306380c. Each thread has its own reply. The sub-issues they touch (#4867, #4873, #4874) were updated to match.

Requesting round 2.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🔵 Needs a closer look

The inventory has missing or inconsistent verdicts and several incorrect dependency relationships.

Review effort: Balanced
Findings: None

Resolved since last review (8)
Previously missed (1)

In code that hasn't changed since last review

Low severity Replace Open decision with a defined verdict

docs/​agent-guidance-inception.md:82

Open decision is outside the three verdicts defined above and does not satisfy #4859's requirement to classify every Inspector element as Transfer, Adapt, or N/A. This rule is still an adaptation whose final policy is deferred to S8.

This issue also appears in the following locations of the same file:

  • line 108
  • line 127
  • line 397
  • line 552

- Contributing is Adapt, with the policy deferred to S8; qualifiers such
  as 'inverted' and 'later' move out of the verdict column
- Size is not an Inspector element, so it moves out of the inventory table
- S6 after S5; S10 also after S2; S12 also after S11 (client-smoke)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 2: no inline comments, and one finding in "Previously missed" (no thread to reply to, so answered here). Fixed in 0df4677.

Requesting round 3.

@cliffhall
cliffhall requested a balanced review from Copilot September 27, 2026 03:39
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 4: one inline finding plus two in "Previously missed". All fixed in a3e0e86.

Requesting round 5.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🔵 Needs a closer look

Several planned quality gates are omitted or not wired into CI, and some inventory details conflict with their source.

Review effort: Balanced
Findings: None

Resolved since last review (1)
Previously missed (4)

In code that hasn't changed since last review

Medium severity Shared dependency version guard remains unenforced

docs/​agent-guidance-inception.md:133

A single npm lockfile can still resolve multiple versions for different workspaces. The current manifests already declare different TypeScript ranges (everything uses ^5.6.2, filesystem uses ^5.8.2, and sequentialthinking uses ^5.3.3), so marking this guard N/A leaves the “one version of a shared devDependency” rule on line 78 unenforced.

This issue also appears on line 134 of the same file.

Medium severity Root guards are missing from CI matrix jobs

docs/​agent-guidance-inception.md:444

The adapted root guards are only included in the root aggregate, while every described CI matrix leg runs the package-only validate. Without a dedicated root guard job, verify:format-coverage, verify:typecheck-coverage, and formatting/lint failures in root files are not actually CI-gated.

This issue also appears on line 453 of the same file.

Medium severity Dependabot backlog is not addressed

docs/​agent-guidance-inception.md:602

Six Dependabot PRs are currently open, but this scope only disables future PR creation and does not close the backlog. The created #4874 therefore uses “No new Dependabot PRs open after merge”; this stronger criterion is not satisfied by the described work and no longer matches the sub-issue.

Low severity Misstates Inspector guidance on paths

docs/​agent-guidance-inception.md:87

The Inspector rule does not ban paths; it says to use them only when a skill is useless outside the matched files and to document the tradeoff. Calling this a direct transfer while summarizing it as “no paths” misstates the source and would make S2 unnecessarily stricter.

…log, paths rule

- S3 adds a root-guards CI job for checks no package leg covers
- verify:dep-lockstep is Adapt: one range per shared TS devDependency
- S13 works down the six open Dependabot PRs
- The skills rule allows paths with a stated trade-off, as the Inspector does

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 5: no inline comments, and four findings in "Previously missed". All fixed in 93b83d3.

  • The shared-dependency version guard was marked N/A. Fixed. verify:dep-lockstep is now Adapt: a smaller guard requiring one range for each shared TS devDependency, or hoisting it to the root (today typescript alone has three ranges). It runs in S3's root-guards job. verify:install-fresh stays N/A, because npm ci already fails on a lockfile mismatch.
  • Root guards were missing from the CI matrix. Fixed. S3 adds a separate root-guards CI job for root-file format/lint, verify:format-coverage, verify:typecheck-coverage and the version guard.
  • The Dependabot backlog wasn't addressed. Fixed. S13 now works down the six open Dependabot PRs, converting each still-needed bump to an issue and closing the PR with a pointer. There are separate acceptance criteria for "no new PRs" and "backlog closed".
  • The Inspector's paths guidance was misstated. Fixed. The rule is now "paths only when a skill is useless outside the matched files, with the trade-off stated", matching the source.

#4863, #4864 and #4874 are updated to match. Requesting round 6.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🟡 Changes recommended

The inventory, skills bootstrap, and action-pinning scope contain unresolved gaps.

Review effort: Balanced
Findings: 1 High severity

Open (1)
Previously missed (4)

In code that hasn't changed since last review

Medium severity File purpose-header requirement is not separately addressed

docs/​agent-guidance-inception.md:76

This row combines the annotated-tree rule with the Inspector's separate requirement that every file have a purpose/rationale header, but the adaptation only addresses the tree. Existing source files such as src/filesystem/index.ts and src/time/src/mcp_server_time/server.py do not follow that header convention, so S1 cannot tell whether to introduce a repository-wide requirement or omit it. Give the header rule its own verdict and either scope the migration or mark it N/A with a reason.

Medium severity Workflow gate script and tests are missing from coverage

docs/​agent-guidance-inception.md:122

The pinned Inspector revision also has scripts/lib/workflow-gate.mjs and its tests. That guard enforces the local-vs-CI split and ensures local:gate remains exactly the lease wrapper, but it is absent from the inventory, reusable templates, and S10 even though #4859 requires every gate script to be covered. Add an Adapt/N/A verdict and carry any applicable invariants into S10/#4871.

Medium severity Empty skills directory breaks the ported verifier

docs/​agent-guidance-inception.md:429

The verifier being ported deliberately fails when .claude/skills/ is empty (no skills found; the skill set is gone). Requiring the default command to pass empty either contradicts the port or permanently removes protection against deleting every skill. Seed S2 with at least one real skill, or define an explicitly temporary bootstrap mode and require its removal when the first skill lands.

Low severity Incorrectly claims every server has a package.json

docs/​agent-guidance-inception.md:78

The repository has only four package.json files under src/; the three Python servers use pyproject.toml. Saying every server has its own package.json makes the N/A rationale factually incorrect. Limit this statement to TypeScript servers while retaining the independent-publication point.

Comment thread docs/agent-guidance-inception.md Outdated
…e pin scope

- The per-file purpose header gets its own row: Adapt, with no bulk
  migration
- package.json applies to the four TS servers only
- workflow-gate.mjs is inventoried (Adapt, S10) and listed as a template
- S2 keeps the verifier's no-skills failure, behind a temporary bootstrap
  allowance that the first skill PR removes
- Action pinning covers jobs whose artifacts credentialed jobs download

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 6: one inline finding plus four in "Previously missed". All fixed in c27e937.

Requesting round 7.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🔵 Needs a closer look

The proposed gates omit Python package builds and introduce an absolute PR rule before conflicting automation is retired.

Review effort: Balanced
Findings: None

Resolved since last review (1)
Previously missed (2)

In code that hasn't changed since last review

Medium severity Issue-reference rule conflicts with issue-less Dependabot PRs

docs/​agent-guidance-inception.md:85

Assigning the unconditional “every PR references an issue” rule to S1 conflicts with the current automation documented in §8: Dependabot continues opening issue-less PRs until S13. That also violates S1's requirement that every rule be true when it merges. Either defer this rule to S13 or make the existing automated dependency PRs an explicit temporary exception that S13 removes.

Medium severity Python validation omits the required package build step

docs/​agent-guidance-inception.md:471

The planned Python validate omits package building, but python.yml currently gates every package with uv build after pyright. Since S10 only chains this validate yet promises to run every CI check, the local gate could pass while CI fails to build a wheel. Include uv build in each per-server validation (or add an explicit Python build stage to S10).

- S1 states 'every PR references an issue' with an explicit temporary
  exception for Dependabot PRs, which S13 removes
- S4's per-server validate includes uv build, which CI already gates

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 7: no inline comments, and two findings in "Previously missed". Both fixed in cb52794.

Requesting round 8.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🔵 Needs a closer look

The roadmap contains inconsistent dependencies and first-skill bootstrap responsibilities that could leave planned acceptance criteria unsatisfied.

Review effort: Balanced
Findings: None

Previously missed (3)

In code that hasn't changed since last review

Medium severity Add S1 dependency or handoff rule for S3 and S4

docs/​agent-guidance-inception.md:407

Wave 2 is declared parallel, but S3 and S4 both require edits to the root AGENTS.md that S1 creates, while neither has an S1 dependency or S2's explicit handoff fallback. If either lands first, it must either create/overlap S1's file or omit part of its scope. Add an S1 dependency or the same handoff rule to S3/S4, and synchronize #4864/#4865.

Medium severity Include S10 and S11 in empty-skills bootstrap removal

docs/​agent-guidance-inception.md:435

This candidate list omits S10 and S11, both of which add skills and are allowed by their declared dependencies to run after S2 without waiting for S5 or S9. If either lands first, the empty-skills bootstrap remains and deleting every skill can still pass verification. Make every potentially first skill PR (S5, S9, S10, and S11) responsible for removing it, and update #4863/#4871/#4872 accordingly.

This issue also appears on line 480 of the same file.

Medium severity Align S11 regression requirement with available skills

docs/​agent-guidance-inception.md:575

S11 only depends on S2, so it may run before the Wave 3 skills exist; requiring a regression run specifically against those skills is then unsatisfiable. Match #4872's acceptance criterion by saying “skills that already exist,” or add explicit Wave 3 dependencies.

…moval

- S3 and S4 hand their AGENTS.md rules to S1 if it hasn't merged yet
- Whichever skill-adding issue lands first (S5, S6, S9, S10 or S11)
  removes S2's empty-skills bootstrap allowance
- S11's regression check targets the skills that already exist

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 8: no inline comments, and three findings in "Previously missed". All fixed in 3bad965.

Requesting round 9.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🟡 Changes recommended

Release security pinning and final documentation are scheduled after steps that already depend on them.

Review effort: Balanced
Findings: 1 High severity

Open (1)
Previously missed (1)

In code that hasn't changed since last review

Low severity Finalize the factory overview after Wave 6 automation

docs/​agent-guidance-inception.md:606

This “closing” factory document is scheduled in Wave 5, but Wave 6 still adds the dependency/security sweeps, SDK watch, action-pin guard, and final AGENTS.md rule changes. Publishing the overview in S12 will therefore describe an incomplete factory and become stale immediately. Move this task to S13, or explicitly require S13 to finalize/update it after automation lands.

Comment thread docs/agent-guidance-inception.md Outdated
- Action pinning, including the jobs that produce artifacts credentialed
  jobs consume, lands with the release split, before the first milestone
  release
- The closing docs/ai-software-factory.md is written in S13, after Wave 6

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review, round 9: one inline finding plus one in "Previously missed". Both fixed in 278a9f2.

Requesting round 10.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot review overview

🟢 Approval recommended

The documentation comprehensively satisfies the linked issue and presents a consistent, actionable implementation plan.

Review effort: Balanced
Findings: None

Resolved since last review (1)

@cliffhall

Copy link
Copy Markdown
Member Author

Copilot review loop closed: round 10 was clean. It had no inline comments and nothing in "Previously missed", and the overview reads "🟢 Approval recommended". Rounds 1–9 raised 29 findings in total. All were fixed, and the affected sub-issues (#4862–#4874) and the #4858 tracker were kept in sync. The PR is ready for maintainer review.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Agentic software factory Part 1: Inception — inventory the Inspector factory, write docs/agent-guidance-inception.md

2 participants