docs(agents): extend the coding-agent catalog to cover agentic platforms (closes #86) (#215)

* start: survey agentic AI platforms (closes #86)

* docs(agents): survey agentic platforms by execution model (closes #86)

Issue #86 asked for a survey of the popular agentic platforms, naming
Cursor, OpenCode, Claude Code, Codex, Antigravity, Jules, and Ollama.
The catalog (#196) and harness landscape (#95) already describe most of
them, but the platform table in coding-agent-platforms.qmd had no row
for OpenCode, Cline, Kiro, Warp, OpenHands, Devin, or Ollama, and
nothing explained the one dimension that decides where a reader's code
goes: whether the agent runs in a hosted sandbox, on the local checkout,
or as an orchestration layer over other agents.

Extend the existing fragments rather than add a parallel survey: add the
missing rows with citations to the vendor pages fetched on 2026-09-09,
add a "Where the agent runs" section that sorts every platform into the
three execution models and notes that windsurf.com now redirects to
Devin Desktop, and mention OpenCode alongside Aider and OpenHands in the
Ollama harness list, since it is the stack the site's own OpenCode
sections document.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* docs(platforms): cite the Devin row and add four wordlist entries

The Spellcheck job failed on `Bitbucket`, `DevOps`, `webhooks`, and
`rebranding`, which the new prose introduced without wordlist entries
(only the singular `webhook` and the past-tense `rebranded` were
listed). The Devin row was the one new platform row without a
citation; it now cites the Devin documentation's introduction page,
fetched 2026-09-09.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* docs(platforms): finish the truncated "token-cost characteristics" phrase

Copilot read "token-cost character" as a truncation, which it was.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* fix(bib): rebuild references.bib as a clean union after the main sync

The union resolver used for the sync dropped the blank line between
every pre-existing entry and truncated one entry, which made pandoc
reject the file and failed the build. The file is now rebuilt from
the two parent versions: main's entries in main's order and formatting,
then this branch's new entries appended, validated with pandoc.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

Author: dem-extra1

chapters/ai-tools/catalog-coding-agents.qmd

@@ -31,7 +31,8 @@ with dedicated agent harnesses (detailed setup and hardware sizing are covered i
 @sec-ai-offline):
 
 - **Terminal Orchestrators**:
-  Harnesses like **[Aider](https://aider.chat/)** and **[OpenHands](https://github.com/All-Hands-AI/OpenHands)**
+  Harnesses like **[OpenCode](https://opencode.ai)** (see @sec-ai-opencode-ollama),
+  **[Aider](https://aider.chat/)**, and **[OpenHands](https://github.com/All-Hands-AI/OpenHands)**
   (formerly OpenDevin) connect directly to Ollama endpoints running open-weight
   coding models (such as `Qwen2.5-Coder`, `DeepSeek-Coder-V2`, or `Llama 3.3`),
   managing git commits, multi-file edits, and automated test-and-repair loops.

chapters/ai-tools/coding-agent-platforms.qmd

@@ -10,13 +10,20 @@ The following catalog is a starting point rather than an endorsement:
 |---|---|---|
 | [GitHub Copilot coding agent](https://github.com/features/copilot/agents) | GitHub issues and pull requests | Assign an issue; review the resulting pull request |
 | [OpenAI Codex](https://openai.com/codex/) | Cloud tasks, app, and command line | Delegate a task in an isolated environment or work locally |
-| [Google Jules](https://jules.google.com/) | Cloud coding agent | Connect a repository and review the proposed changes |
+| [Google Jules](https://jules.google.com/) [@jules_docs] | Cloud coding agent | Connect a repository and review the proposed changes |
 | [Google Antigravity](https://antigravity.google/) | Agentic development platform | Coordinate coding tasks in a managed development workspace |
 | [Claude Code](https://www.anthropic.com/claude-code) | Terminal, IDE, and cloud | Work interactively or delegate work that returns a pull request |
 | [Gemini CLI](https://github.com/google-gemini/gemini-cli) | Command line | Inspect, edit, and test the current local checkout |
-| [Cursor](https://www.cursor.com/) | IDE | Edit interactively with path-scoped `.cursor/rules` context |
+| [Cursor](https://www.cursor.com/) [@cursor_cloud_agents] | IDE and cloud agents | Edit interactively with path-scoped `.cursor/rules` context, or hand a task to a cloud agent that pushes a branch |
 | [Aider](https://aider.chat/) | Command line | Pair locally with explicit files and Git commits |
 | [GitKraken Kepler](https://gitkraken.com/kepler) | Agentic development environment (desktop) | Start a Task from an issue, PR, or idea; run multiple agents in parallel isolated worktrees via reusable Actions; review per-branch diffs and open PRs |
+| [OpenCode](https://opencode.ai) [@opencode_docs] | Terminal, desktop app, and IDE extension | Bring your own model provider (including local endpoints; see @sec-ai-opencode-ollama) and edit the local checkout |
+| [Cline](https://cline.bot/) [@cline_site] | VS Code extension, CLI, and SDK | Plan, then act with per-step approval; bring your own key, including Ollama |
+| [Kiro](https://kiro.dev/) [@kiro_docs] (AWS) | IDE, CLI, web, and mobile | Spec-driven tasks locally, or delegate a web task that opens a pull request |
+| [Warp](https://docs.warp.dev/) [@warp_docs] | Terminal app, standalone CLI, and cloud agents | Approve actions locally, or trigger cloud agents from Slack, Linear, or GitHub webhooks |
+| [OpenHands](https://docs.openhands.dev/) [@openhands_docs] | Browser client, managed cloud, self-hosted backend, and CLI | Run headless or hosted; self-host when code must stay on your infrastructure |
+| [Devin](https://devin.ai/) [@devin_docs] (Cognition) | Cloud agent and desktop editor | Delegate a task to an isolated cloud machine and review the pull request (see @sec-ai-harness-landscape) |
+| [Ollama](https://ollama.com/) [@ollama_site] | Local model runner (not an agent) | Serve open-weight models to any of the harnesses above (see @sec-ai-ollama-agents) |
 
 Kepler is not an agent itself but an orchestration layer.
 It hosts agents you already use
@@ -27,6 +34,55 @@ Where a single-agent platform handles one repo at a time,
 a Kepler Task can span many repos,
 and its Agent Graph visualizes every session, turn, tool call, and subagent live.
 
+#### Where the agent runs {#sec-ai-platform-execution-models}
+
+The platforms above sort into three execution models
+(measured 2026-09-09).
+The model decides where your code is copied,
+what the agent can reach,
+and how its work comes back to you.
+
+- **Hosted sandbox.**
+  The vendor provisions an isolated machine,
+  clones the repository into it,
+  and returns a branch or pull request.
+  Jules clones the repository into a virtual machine and submits a pull request after you approve its plan [@jules_docs];
+  Cursor cloud agents clone from GitHub, GitLab, Azure DevOps, or Bitbucket,
+  work on a separate branch, and push it back [@cursor_cloud_agents];
+  Kiro web tasks and Warp cloud agents follow the same shape [@kiro_docs; @warp_docs],
+  as do the Copilot coding agent, Codex cloud tasks, Devin, and OpenHands Cloud.
+  Nothing runs on your machine,
+  so the questions are what the sandbox can reach
+  and whether the vendor's retention terms suit the repository.
+- **Local checkout.**
+  The agent runs on your workstation against the files already there,
+  in a terminal (Claude Code, Codex CLI, OpenCode, Aider, Gemini CLI, Warp)
+  or inside an editor (Cursor, Cline, Kiro, Copilot in VS Code).
+  Code stays put,
+  and the agent inherits whatever credentials and network access your shell has,
+  which is why @sec-ai-best-practices asks for approval gates and a sandbox.
+  Pairing one of these with Ollama keeps the model local too [@ollama_site],
+  at the cost of the hardware described in @sec-ai-offline.
+- **Orchestration layer.**
+  Kepler, Kiro Crew, and the Cline Kanban surface do not add a model of their own;
+  they run several of the agents above in parallel worktrees or sessions
+  and present the results for review.
+  Reach for one only once a single agent is no longer the bottleneck
+  ([when orchestration helps](agent-orchestration.qmd#sec-orch-when)).
+
+Several vendors now ship all three,
+so the platform name alone no longer tells you where the code goes:
+Claude Code, Codex, Cursor, Kiro, and Warp each offer a local surface and a hosted one.
+Check the execution model of the specific surface you enable.
+The Windsurf editor is a naming trap of the same kind:
+`windsurf.com` now redirects to Devin Desktop,
+Cognition's rebranding of that editor (measured 2026-09-09; see @sec-ai-harness-landscape).
+
+For licensing, token-cost characteristics, and a decision table for choosing among these tools,
+see @sec-ai-harness-landscape;
+for the open-weight and fully local end of the spectrum,
+see @sec-ai-catalog-coding-agents.
+
 Before connecting any platform, check:
 
 - whether code runs locally or in a vendor-managed environment;

inst/WORDLIST

@@ -31,6 +31,7 @@ Battlestar
 Belcak
 BenchLM
 Bing
+Bitbucket
 Bot's
 CAPTCHAs
 CHANGELOG
@@ -75,6 +76,8 @@ Daytona
 DeepSeek
 DeepSeek-R1
 DeepSeekMath
+DevOps
+Devin
 Devstral
 Docker
 DuckDB
@@ -145,6 +148,7 @@ Karpathy's
 Keanu
 Kepler
 Kimi
+Kiro
 Kong
 Kotlin
 Kubrick
@@ -316,6 +320,7 @@ Wachowskis
 Wakatime
 WhatsApp
 Wiggum
+Windsurf
 Workspaces
 YAGNI
 YAML
@@ -503,6 +508,7 @@ rStar
 reachability
 rebase
 rebranded
+rebranding
 reflow
 refspec
 renderer
@@ -589,6 +595,7 @@ wakeup
 wakeup's
 walkthrough
 webhook
+webhooks
 workspace's
 workspaces
 workstreams

references.bib

@@ -1186,3 +1186,84 @@ @misc{perplexity_sonar_deep_research
   url = {https://docs.perplexity.ai/docs/sonar/models/sonar-deep-research},
   note = {Accessed 2026-09-09}
 }
+
+@misc{opencode_docs,
+  title = {Intro: AI coding agent built for the terminal},
+  author = {{OpenCode}},
+  year = {2026},
+  howpublished = {Documentation},
+  url = {https://opencode.ai/docs/},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{cline_site,
+  title = {Cline: AI Coding, Open Source and Open Choice},
+  author = {{Cline}},
+  year = {2026},
+  howpublished = {Product page},
+  url = {https://cline.bot/},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{kiro_docs,
+  title = {Kiro Documentation},
+  author = {{Amazon Web Services}},
+  year = {2026},
+  howpublished = {Documentation},
+  url = {https://kiro.dev/docs/},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{warp_docs,
+  title = {Getting started with Warp},
+  author = {{Warp}},
+  year = {2026},
+  howpublished = {Documentation},
+  url = {https://docs.warp.dev/},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{openhands_docs,
+  title = {OpenHands documentation: Introduction},
+  author = {{OpenHands}},
+  year = {2026},
+  howpublished = {Documentation},
+  url = {https://docs.openhands.dev/},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{cursor_cloud_agents,
+  title = {Cloud Agents},
+  author = {{Cursor}},
+  year = {2026},
+  howpublished = {Documentation},
+  url = {https://cursor.com/docs/cloud-agent},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{jules_docs,
+  title = {Getting started with Jules},
+  author = {{Google}},
+  year = {2026},
+  howpublished = {Documentation},
+  url = {https://jules.google.com/docs/},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{ollama_site,
+  title = {Ollama},
+  author = {{Ollama}},
+  year = {2026},
+  howpublished = {Product page},
+  url = {https://ollama.com/},
+  note = {Accessed 2026-09-09}
+}
+
+@misc{devin_docs,
+  title = {Introducing Devin},
+  author = {{Cognition}},
+  year = {2026},
+  howpublished = {Documentation},
+  url = {https://docs.devin.ai/get-started/devin-intro},
+  note = {Accessed 2026-09-09}
+}