diff --git a/docs/seo-strategy/citation-deltas/README.md b/docs/seo-strategy/citation-deltas/README.md new file mode 100644 index 0000000..a43f12d --- /dev/null +++ b/docs/seo-strategy/citation-deltas/README.md @@ -0,0 +1,58 @@ +# Citation Deltas — Inline Citation Pass for Published Posts + +Research artifact from issue #392 (Category J of the SEO + AI-discovery plan). + +The 4 published blog posts have zero inline citations despite narratively referencing Anthropic essays, arXiv papers, named researchers, and framework docs. Inline citations are associated with higher AI citation lift in generative-engine-optimization research (GEO, KDD 2024). This directory proposes specific link insertions for Julian's manual review and application via Payload admin. + +Nothing here modifies post bodies. Nothing here touches production code. Every proposed URL was verified to return HTTP 200 (or an equivalent successful response) at the time of writing. + +## Posts covered + +| Post | Delta file | Proposed citations | +| ------------------------------------------ | -------------------------------------------------------------------------------- | ------------------ | +| `where-agentic-patterns-actually-live` | [where-agentic-patterns-actually-live.md](./where-agentic-patterns-actually-live.md) | 8 | +| `subagent-orchestration-workflow` | [subagent-orchestration-workflow.md](./subagent-orchestration-workflow.md) | 2 | +| `what-tickets-and-prs-are-actually-for` | [what-tickets-and-prs-are-actually-for.md](./what-tickets-and-prs-are-actually-for.md) | 1 | +| `rethinking-systems-in-the-agentic-age` | [rethinking-systems-in-the-agentic-age.md](./rethinking-systems-in-the-agentic-age.md) | 1 | + +Total: 12 proposed citations across 4 posts. + +## Application workflow + +For each delta: + +1. Open the Payload admin at `https://detached-node.dev/admin` (or local equivalent). +2. Navigate to **Posts** and select the post by slug. +3. In the Lexical body editor, locate the passage cited in the delta — search by the original snippet's distinctive phrasing. +4. Select the proposed anchor text in the editor. +5. Open the link drawer (toolbar button or `Cmd-K`). Paste the proposed URL. Confirm. +6. Repeat for all proposed citations in the delta. +7. Save and re-publish. + +### Lexical link drawer gotcha + +Programmatic URL setting (e.g., `react-hook-form` `setValue`, or `evaluate_script` injection) drops the value silently. Use real keystrokes — manual paste-then-type by hand, or simulated `browser_type` if automated. The reference memory `reference_payload_lexical_links_via_playwright.md` documents this. + +## Source verification + +Every URL in these deltas was verified to return a successful response at the time of writing (2026-05-17). Re-verify before applying if significant time has passed; the Anthropic engineering essays in particular have moved between `www.anthropic.com/engineering/...` and `code.claude.com/docs/en/...` once already in 2025-2026. + +When a redirect is involved, the delta lists the final destination URL — not the historical one — so the link is canonical at insertion time. + +## What was deliberately NOT proposed + +- **Author's own analysis and assertions.** Lines like "The named patterns are vocabulary. The work happens one level down" are Julian's voice, not external references; linking them out would be wrong. +- **References where no authoritative source could be confidently identified.** Where a phrase like "an earlier piece on subagent orchestration" could refer to multiple posts or an unpublished draft, the delta notes the ambiguity instead of fabricating a link. +- **General programming concepts.** "Autoregressive", "context window", "pair programming" are common knowledge and do not need citations on a blog post. +- **Internal cross-links.** A pass adding `[earlier piece](/posts/subagent-orchestration-workflow)` style internal links is a separate concern (related-posts surface, internal-link graph) and out of scope for this delta set. + +## How to read a delta entry + +Each citation entry has four parts: + +1. **Original snippet** — three or so lines of verbatim context around the reference, so Julian can find the passage in the editor. +2. **Proposed edit** — the same snippet with `[anchor text](url)` inserted at the natural reference point. The anchor text is bolded as the link itself; surrounding prose is untouched. +3. **Source** — the verified URL. +4. **Rationale** — one sentence explaining why this source and which discovery audience benefits (AI citation, organic search, employer, recruiter, peer engineers). + +A short "References without confidently identifiable sources" section at the end of each delta lists anything Julian should disambiguate himself before applying the rest. diff --git a/docs/seo-strategy/citation-deltas/rethinking-systems-in-the-agentic-age.md b/docs/seo-strategy/citation-deltas/rethinking-systems-in-the-agentic-age.md new file mode 100644 index 0000000..ac0bab7 --- /dev/null +++ b/docs/seo-strategy/citation-deltas/rethinking-systems-in-the-agentic-age.md @@ -0,0 +1,38 @@ +# Citation Deltas — `rethinking-systems-in-the-agentic-age` + +Post URL: https://detached-node.dev/posts/rethinking-systems-in-the-agentic-age +Published: April 19, 2026 + +## Summary + +- **Proposed citations:** 1 +- **Citations confirmed verified (HTTP 200 / successful fetch):** 1 +- **References mentioned but no authoritative source found:** 0 + +This is the earliest published post and is largely a posture-setting essay — what to evolve, what to leave behind, where to spend attention, how to work together. The post deliberately stays at the level of framing, so it names very few concrete external technologies. The one inline citation worth adding is to MCP, which is the only named protocol in the body. + +## Proposed insertions + +### Citation 1: MCP / connectors (the only named protocol in the post) + +**Original snippet:** + +> They were created and adopted to solve specific problems, and those problems are not just going to disappear. It will take time for our ways of working to evolve, and many of these technologies may be able to evolve along with us. Properly maintained connectors or MCP servers might be enough for us to continue using these services as-is. + +**Proposed edit:** + +> They were created and adopted to solve specific problems, and those problems are not just going to disappear. It will take time for our ways of working to evolve, and many of these technologies may be able to evolve along with us. Properly maintained connectors or [MCP servers](https://modelcontextprotocol.io/) might be enough for us to continue using these services as-is. + +**Source:** https://modelcontextprotocol.io/ +**Source verified:** YES on 2026-05-17. Page titled "What is the Model Context Protocol (MCP)?", canonical home of the open standard. +**Rationale:** MCP is the only specific external technology the post names by name. Linking it grounds the central claim that existing project-management tooling, git providers, and observability platforms may not need to be thrown out wholesale — they may instead be wrapped with MCP. This citation serves the AI-discovery audience (this is the load-bearing technology mention for "agentic integration" queries), the engineer-reader (concrete protocol to follow), and the employer/peer audiences (signals that the author tracks the current agentic stack, not just abstract framing). + +## References without confidently identifiable sources + +None for this post. + +## Note on absent citations + +The post deliberately operates at a posture-setting level — what to evolve, what to leave behind, where to spend attention. References like "AI generated document that the person who generated it hadn't even bothered to read yet", or "the way we've always measured tool effectiveness" are general industry observations, not citations to specific work. Forcing speculative citations into those passages would weaken the post. + +If a future revision wanted one additional citation to strengthen AI-discovery surface area, the natural candidate would be a productivity-measurement source (e.g., DORA, SPACE framework) to ground the line *"Productivity shouldn't be measured by the number or size of the artifacts you generate."* This is flagged here for Julian's awareness but not proposed as a delta, because the post's own framing does not actually claim to be in dialogue with any specific productivity-measurement literature, and force-fitting a DORA citation would over-claim. diff --git a/docs/seo-strategy/citation-deltas/subagent-orchestration-workflow.md b/docs/seo-strategy/citation-deltas/subagent-orchestration-workflow.md new file mode 100644 index 0000000..f9868a2 --- /dev/null +++ b/docs/seo-strategy/citation-deltas/subagent-orchestration-workflow.md @@ -0,0 +1,46 @@ +# Citation Deltas — `subagent-orchestration-workflow` + +Post URL: https://detached-node.dev/posts/subagent-orchestration-workflow +Published: April 24, 2026 + +## Summary + +- **Proposed citations:** 2 +- **Citations confirmed verified (HTTP 200 / successful fetch):** 2 +- **References mentioned but no authoritative source found:** 0 + +This post is mostly Julian's own framing of the orchestrator + clean-context subagent pattern. The two external references that warrant inline citations are: (1) Claude Code as the named tool, and (2) the autoregressive-LLM technical claim, which is grounded enough to deserve a primary source. + +## Proposed insertions + +### Citation 1: Claude Code (tool reference) + +**Original snippet:** + +> A walkthrough of an orchestration pattern for agentic coding: a single orchestrator dispatches scoped subagents to independent tasks, and an independent reviewer inspects each result. The examples use Claude Code, but the concepts apply to any modern agentic coding tool. + +**Proposed edit:** + +> A walkthrough of an orchestration pattern for agentic coding: a single orchestrator dispatches scoped subagents to independent tasks, and an independent reviewer inspects each result. The examples use [Claude Code](https://code.claude.com/docs/en/overview), but the concepts apply to any modern agentic coding tool. + +**Source:** https://code.claude.com/docs/en/overview +**Source verified:** YES on 2026-05-17. The legacy `docs.claude.com/en/docs/claude-code/overview` 301-redirects to `code.claude.com/docs/en/overview`. The destination is the official Claude Code overview page. +**Rationale:** Claude Code is the named tool the entire walkthrough is grounded in, and it appears in the first paragraph without an anchor. Linking out from the first product mention is the cheapest standard SEO/citation hygiene move, and AI overviews summarizing this post will route through this URL when the post is the citation source. Engineer-reader, AI-citation, and recruiter audiences all benefit. + +### Citation 2: Autoregressive LLM behavior (Claude Code best-practices on context-window behavior) + +**Original snippet:** + +> The large language model powering these tools is autoregressive; every new token attends to the full conversation history. As the context grows with each completed task, irrelevant work from earlier tasks is still present and influencing the model's output. + +**Proposed edit:** + +> The large language model powering these tools is autoregressive; every new token attends to the full conversation history. As the context grows with each completed task, [irrelevant work from earlier tasks is still present and influencing the model's output](https://code.claude.com/docs/en/best-practices). + +**Source:** https://code.claude.com/docs/en/best-practices +**Source verified:** YES on 2026-05-17. Page titled "Best practices for Claude Code". Contains the explicit statement: *"LLM performance degrades as context fills. When the context window is getting full, Claude may start 'forgetting' earlier instructions or making more mistakes."* This is the operationally-grounded primary source for the claim Julian is making. +**Rationale:** The autoregressive claim is technical and could be sourced to a transformer paper (Vaswani et al., 2017), but the *operational* claim Julian is making — that residual context degrades subsequent task quality — is better sourced to Claude Code's own best-practices doc, which says exactly this. This citation reinforces the post's central argument about why fresh-context subagents work and routes AI overviews to a load-bearing primary source rather than an academic detour. + +## References without confidently identifiable sources + +None for this post. The remainder of the prose ("This also changes how review works...", "To structure your work, you will have subagents create tickets...") is Julian's own framing of the orchestration pattern and is correctly unattributed. diff --git a/docs/seo-strategy/citation-deltas/what-tickets-and-prs-are-actually-for.md b/docs/seo-strategy/citation-deltas/what-tickets-and-prs-are-actually-for.md new file mode 100644 index 0000000..05c77c3 --- /dev/null +++ b/docs/seo-strategy/citation-deltas/what-tickets-and-prs-are-actually-for.md @@ -0,0 +1,36 @@ +# Citation Deltas — `what-tickets-and-prs-are-actually-for` + +Post URL: https://detached-node.dev/posts/what-tickets-and-prs-are-actually-for +Published: April 20, 2026 + +## Summary + +- **Proposed citations:** 1 +- **Citations confirmed verified (HTTP 200 / successful fetch):** 1 +- **References mentioned but no authoritative source found:** 0 + +This post is almost entirely Julian's own argument about why ticketing and PR conventions still pay rent in an agentic workflow. The only external reference that benefits from an inline citation is the MCP / connector mention, which references a concrete open standard with a canonical URL. + +## Proposed insertions + +### Citation 1: MCP / connectors (the protocol the prose names directly) + +**Original snippet:** + +> The same artifacts are useful to an AI agent. Dispatching a fresh agent to pick up a ticket lets it start with a clean context and zero in on what matters for execution. With properly set up connectors or MCP servers, an agent can: + +**Proposed edit:** + +> The same artifacts are useful to an AI agent. Dispatching a fresh agent to pick up a ticket lets it start with a clean context and zero in on what matters for execution. With properly set up connectors or [MCP servers](https://modelcontextprotocol.io/), an agent can: + +**Source:** https://modelcontextprotocol.io/ +**Source verified:** YES on 2026-05-17. Page titled "What is the Model Context Protocol (MCP)?", canonical home of the open standard. +**Rationale:** MCP is the specific protocol the prose names. It is the only concrete external technology in the entire post, and linking it grounds the agent-readability claim that follows (traverse linked docs, interpret design mocks, ingest prior decisions). Linking the protocol home rather than the Claude Code MCP doc keeps the citation tool-agnostic, matching the post's tool-neutral framing. Serves AI-discovery (queries about agent + ticketing integration), engineer-reader (concrete protocol to investigate), and employer/peer audiences (signals literacy in current agentic stack). + +## References without confidently identifiable sources + +None for this post. The other references the post makes — to "convergence over decades", to engineering teams that "discovered their value the hard way", to "engineering team norms" — are general claims about the practice of software engineering, not citations to specific named work, and should remain unlinked. Adding speculative citations there would weaken rather than strengthen the post's authority. + +## Note on absent citations + +The post argues a position about the durability of ticketing and PR norms in agentic workflows. It is correct that this is largely original argument, not a literature survey, and the lack of additional citations is by design. The post's discoverability gains will come mostly from the named MCP citation above plus internal cross-links (e.g., to `subagent-orchestration-workflow` for the "multiple lenses" PR-review framing) — but internal cross-linking is out of scope here (see README). diff --git a/docs/seo-strategy/citation-deltas/where-agentic-patterns-actually-live.md b/docs/seo-strategy/citation-deltas/where-agentic-patterns-actually-live.md new file mode 100644 index 0000000..eddee4b --- /dev/null +++ b/docs/seo-strategy/citation-deltas/where-agentic-patterns-actually-live.md @@ -0,0 +1,141 @@ +# Citation Deltas — `where-agentic-patterns-actually-live` + +Post URL: https://detached-node.dev/posts/where-agentic-patterns-actually-live +Published: May 15, 2026 + +> **Note on grouping:** Although 8 citations are proposed below, they group into 4 unique anchor passages, not 8 independent edit cycles: +> +> - **Passage 1** (Anthropic essay sentence) — Citation 1 +> - **Passage 2** (Reflexion + Self-Refine sentence) — Citations 2 and 3 land together +> - **Passage 3** (Anthropic's SDK docs sentence) — Citation 4 +> - **Passage 4** (subagents / skills / hooks / MCP servers sentence) — Citations 5, 6, 7, 8 land together +> +> The README workflow describes per-citation cycles, but Citations 3 and 6-8 are marked "(Combined with Citation 2/5 above.)" because they share an anchor passage. Plan **4 edit cycles** in the Payload admin, one per passage, each cycle inserting all the inline links for that passage at once — not 8 separate cycles per individual link. + +## Summary + +- **Proposed citations:** 8 +- **Citations confirmed verified (HTTP 200 / successful fetch):** 8 +- **References mentioned but no authoritative source found:** 1 + +This post is the densest source of unattributed external references in the published set. It names two arXiv papers, the Anthropic "Building effective agents" essay, Claude Code primitives (subagents, skills, hooks, MCP), and the superpowers plugin's writing-skills skill — none of which currently link out. + +## Proposed insertions + +### Citation 1: Anthropic's December 2024 "Building effective agents" essay + +**Original snippet:** + +> The shared names mostly come from Anthropic's December 2024 "Building effective agents" post, which lists six workflows: prompt chaining, routing, parallelization with sectioning and voting, orchestrator-workers, evaluator-optimizer, and the open-ended "agents" category. + +**Proposed edit:** + +> The shared names mostly come from Anthropic's December 2024 ["Building effective agents"](https://www.anthropic.com/engineering/building-effective-agents) post, which lists six workflows: prompt chaining, routing, parallelization with sectioning and voting, orchestrator-workers, evaluator-optimizer, and the open-ended "agents" category. + +**Source:** https://www.anthropic.com/engineering/building-effective-agents +**Source verified:** YES on 2026-05-17. Confirmed as Anthropic's December 19, 2024 engineering post by that title. +**Rationale:** This is the load-bearing taxonomy citation for the whole post — the entire "shared names" framing is grounded in this essay. The page is the primary AI-discovery target for queries about agentic workflow patterns, so anchoring directly to it both grounds the post and surfaces it in retrievals that follow this citation. + +### Citation 2: Reflexion paper (Shinn et al., 2023) + +**Original snippet:** + +> Reflection isn't on that list; it traces back to the Reflexion paper (Shinn et al., 2023) and Self-Refine (Madaan et al., 2023). I'll flag jargon where I use it. + +**Proposed edit:** + +> Reflection isn't on that list; it traces back to the [Reflexion paper (Shinn et al., 2023)](https://arxiv.org/abs/2303.11366) and [Self-Refine (Madaan et al., 2023)](https://arxiv.org/abs/2303.17651). I'll flag jargon where I use it. + +**Source:** https://arxiv.org/abs/2303.11366 +**Source verified:** YES on 2026-05-17. Confirmed title "Reflexion: Language Agents with Verbal Reinforcement Learning", authors include Noah Shinn et al., v4 submitted October 10, 2023. +**Rationale:** Academic citation in the canonical place — arXiv abstract page. Reflexion is one of two named papers in the post and the standard reference for the verbal-reflection pattern; linking it raises the page's authority signal for AI overviews summarizing reflection patterns, and serves engineer-readers who want to chase the primary source. + +### Citation 3: Self-Refine paper (Madaan et al., 2023) + +**Original snippet:** + +> Reflection isn't on that list; it traces back to the Reflexion paper (Shinn et al., 2023) and Self-Refine (Madaan et al., 2023). I'll flag jargon where I use it. + +**Proposed edit:** + +> (Combined with Citation 2 above.) Reflection isn't on that list; it traces back to the [Reflexion paper (Shinn et al., 2023)](https://arxiv.org/abs/2303.11366) and [Self-Refine (Madaan et al., 2023)](https://arxiv.org/abs/2303.17651). I'll flag jargon where I use it. + +**Source:** https://arxiv.org/abs/2303.17651 +**Source verified:** YES on 2026-05-17. Confirmed title "Self-Refine: Iterative Refinement with Self-Feedback", lead author Aman Madaan, submitted March 30, 2023. +**Rationale:** Paired companion to Reflexion in the post's own phrasing. Same rationale: academic peer-source for the reflection pattern, picks up the AI-discovery audience that follows up on reflection lineage and the engineer-reader who reads primary papers. + +### Citation 4: Anthropic SDK docs (the "build from scratch" reference) + +**Original snippet:** + +> Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover subagents, skills, hooks, and MCP servers individually. + +**Proposed edit:** + +> [Anthropic's SDK docs](https://github.com/anthropics/anthropic-sdk-python) show you how to build the patterns from scratch in code. The feature docs cover subagents, skills, hooks, and MCP servers individually. + +**Source:** https://github.com/anthropics/anthropic-sdk-python +**Source verified:** YES on 2026-05-17. Confirmed as the official `anthropics/anthropic-sdk-python` repo — the canonical Python client library for the Anthropic API, with `examples/` covering tool use, streaming, and message handling patterns developers compose into agent workflows. +**Rationale:** Preserves Julian's existing prose ("Anthropic's SDK docs") and points it at the actual SDK rather than the Cookbook. The post's contrast is between the SDK as a *library/API for building patterns programmatically* and the *feature docs* that cover the four primitives individually — that's an SDK-vs-features framing, not a cookbook-vs-features framing. The Python SDK is the more visible of the two (Python and TypeScript SDKs sit side by side); either would be defensible, but the Python repo is the more common landing surface for "Anthropic SDK" queries. The Cookbook, while topically adjacent, is explicitly a *pattern examples* repo — labeling it "Anthropic's SDK docs" would be inaccurate. + +### Citation 5: Claude Code subagents docs + +**Original snippet:** + +> Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover subagents, skills, hooks, and MCP servers individually. + +**Proposed edit:** + +> Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover [subagents](https://code.claude.com/docs/en/sub-agents), [skills](https://code.claude.com/docs/en/skills), [hooks](https://code.claude.com/docs/en/hooks), and [MCP servers](https://modelcontextprotocol.io/) individually. + +**Source:** https://code.claude.com/docs/en/sub-agents +**Source verified:** YES on 2026-05-17. The previous `docs.claude.com/en/docs/claude-code/sub-agents` path now 301-redirects to `code.claude.com/docs/en/sub-agents`; the final destination resolves and is titled "Create custom subagents". +**Rationale:** This is the primary canonical doc for one of the four primitives the post is explicitly bridging onto. Linking each named primitive to its own doc serves the engineer-reader who needs the actual feature reference, and reinforces topical authority for "Claude Code subagents" AI queries. + +### Citation 6: Claude Code skills docs + +**Original snippet:** + +> Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover subagents, skills, hooks, and MCP servers individually. + +**Proposed edit:** + +> (Combined with Citation 5 above.) Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover [subagents](https://code.claude.com/docs/en/sub-agents), [skills](https://code.claude.com/docs/en/skills), [hooks](https://code.claude.com/docs/en/hooks), and [MCP servers](https://modelcontextprotocol.io/) individually. + +**Source:** https://code.claude.com/docs/en/skills +**Source verified:** YES on 2026-05-17. Page titled "Extend Claude with skills", documents SKILL.md format, frontmatter fields, and skill-content lifecycle. +**Rationale:** Same as Citation 5 — the canonical doc for one of the four named primitives. The skills page is also where the SKILL.md format is defined, which the post directly references later ("A more long-term solution would be encoding that prompt as a SKILL.md file..."). + +### Citation 7: Claude Code hooks docs + +**Original snippet:** + +> Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover subagents, skills, hooks, and MCP servers individually. + +**Proposed edit:** + +> (Combined with Citation 5 above.) Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover [subagents](https://code.claude.com/docs/en/sub-agents), [skills](https://code.claude.com/docs/en/skills), [hooks](https://code.claude.com/docs/en/hooks), and [MCP servers](https://modelcontextprotocol.io/) individually. + +**Source:** https://code.claude.com/docs/en/hooks +**Source verified:** YES on 2026-05-17. Page titled "Hooks reference", documents the SubagentStop hook the post specifically names in the next section ("A SubagentStop hook (or a convention inside the subagent file) appends a structured note..."). +**Rationale:** Same as Citations 5-6, with extra justification: the post later names `SubagentStop` specifically, and the hooks reference is the exact page that documents that hook event. Linking up front primes the reader for the named-hook reference that follows. + +### Citation 8: Model Context Protocol home + +**Original snippet:** + +> Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover subagents, skills, hooks, and MCP servers individually. + +**Proposed edit:** + +> (Combined with Citation 5 above.) Anthropic's SDK docs show you how to build the patterns from scratch in code. The feature docs cover [subagents](https://code.claude.com/docs/en/sub-agents), [skills](https://code.claude.com/docs/en/skills), [hooks](https://code.claude.com/docs/en/hooks), and [MCP servers](https://modelcontextprotocol.io/) individually. + +**Source:** https://modelcontextprotocol.io/ +**Source verified:** YES on 2026-05-17. Page titled "What is the Model Context Protocol (MCP)?", the canonical home of the open standard. +**Rationale:** MCP is its own open standard, not strictly an Anthropic feature — the canonical home is `modelcontextprotocol.io`, not Claude Code's docs. Linking the protocol itself rather than Claude Code's MCP integration page is more accurate to what the post says ("MCP servers" generally, not "Claude Code's MCP support"), and serves the broader audience that lands on this post via MCP-flavored queries. + +## References without confidently identifiable sources + +- **"the superpowers plugin"** in the line *"writing-skills in the superpowers plugin is the one I reach for"*. The most likely target is `github.com/obra/superpowers` (confirmed as the official superpowers plugin for Claude Code, containing the writing-skills skill among others). However, Julian may have intended a different distribution channel — the plugin marketplace listing, his own fork, or a more specific path inside the repo. Recommended URL if Julian confirms: `https://github.com/obra/superpowers`, optionally narrowed to `https://github.com/obra/superpowers/tree/main/skills/writing-skills` if such a directory exists at apply time. Flagged here rather than included in the count of 8 because Julian should disambiguate his preferred surface. + +- **"an earlier piece on subagent orchestration"** in the Critic section. This is plausibly an internal link to `/posts/subagent-orchestration-workflow`. Internal cross-linking is out of scope for this delta (see README — internal-link audit is a separate concern), but flagging it here so Julian can decide whether to add it during the same Payload edit pass.