Skip to content
approved Visibility internal Owner erik@uvilo.com Approver _ Created 2026-07-17 Updated 2026-07-18

Knowledge Consistency Research


Sources Reviewed

Knowledge Files (Forge root)

FileStatus
Forge/AGENTS.mdpublished
Forge/README.mdpublished
Forge/Forge_Charter.mdapproved
Forge/Forge_Agent_Architecture.mdapproved
Forge/Forge_Agent_Orchestration.mdpublished
Forge/Forge_Agents_User_Guide.mdpublished
Forge/Forge_Content_System.mdapproved
Forge/Forge_Document_Structure.mdapproved
Forge/Forge_Infrastructure.mdapproved
Forge/Forge_Markdown_MDX_Syntax.mdxpublished
Forge/Forge_Markdown_Syntax.mdpublished
Forge/Forge_Setup.mdapproved

Knowledge Files (Forge_Project)

FileStatus
Forge/Forge_Project/README.mdpublished
Forge/Forge_Project/AGENTS.mdmissing
Forge/Forge_Project/Forge_Project_Workflow.mdpublished
Forge/Forge_Project/Forge_Project_User_Guide.mdpublished

Bot System Prompts (forge group)

BotHandleModel
Forge Chatchatglm-5.2
Forge Project Runnerproject-runnerglm-5.2
Forge Task Runnertask-runner(not read — no unique prompt)
Forge Project Thinkerproject-thinkerglm-5.2
Forge Project Workerproject-workerglm-5.2
Forge Project Evaluatorproject-evaluatorgpt-5.5
Forge Page Managerpage-managerglm-5.2

Skills (Forge/Skills/ — 26 SKILL.md files)

All 26 skills reviewed: Artifacts_Markdown, Audit_Context, Build_Typescript, Context_Hygiene, Create_New_Version, Create_Skill, Forge_Optimizer, Git_Dashboard, Home_Page, Linear_MCP, MCP_Development, Manage_Pages, Markdown_To_PDF, NPM, Orchestration, Playwright_MCP, Preview, Reindex_Typesense, Start_Astro_Dev, Suprsend_MCP, Test_Tools, Typesense_MCP, Update_Infrastructure, Vercel_MCP, Verify_Infrastructure, Write_Typescript.

Charter

Forge/Forge_Charter.md — read in full.


Category 1: Deprecated

1.1 Orchestration Skill Is Deprecated But Still Listed

R2: Deprecated Items

Finding: Forge/Skills/Orchestration/SKILL.md explicitly states: “This Skill is deprecated. Use the prompts in /references as references for building skills.” However, orchestration is still listed in the Available Skills list in the Persistent Knowledge shared prompt (attached to all 7 forge-group bots).

Source: Forge/Skills/Orchestration/SKILL.md (frontmatter: status: published); Persistent Knowledge prompt (prm_cmq83857c000101mzrhmopc03).

Recommendation: Either remove orchestration from the Available Skills list in the Persistent Knowledge prompt, or update the skill to remove the deprecation notice if the references are still useful. If the skill folder is retained for reference, change its status to archived and remove it from the Available Skills list.

Decision: Remove deprecated status. Explain that references contain system prompts for OpenClaw and Clause Code that can be used as inspiration and information. We also need to rename this to something like “Agentic Prompts”


1.2 Update_Infrastructure Skill Is Marked Outdated But Still Listed

R2: Deprecated Items

Finding: Forge/Skills/Update_Infrastructure/SKILL.md states: ”> This Skill is currently outdated. Do not use!” Yet update-infrastructure remains in the Available Skills list in the Persistent Knowledge prompt, and the skill’s frontmatter status is published.

Source: Forge/Skills/Update_Infrastructure/SKILL.md; Persistent Knowledge prompt.

Recommendation: Either update the skill to reflect current infrastructure practices, or change its status to archived and remove it from the Available Skills list. The Verify_Infrastructure skill already covers post-setup verification.

Decision: Update the skill to reflect current infrastructure practices. You can refer to {prodRepoRoot}/docs/deployment.md if it help (and the rest of the uvilo-mono repo.


1.3 Forge_Chat_Prompt.md Referenced But Does Not Exist

R2: Deprecated Items

Finding: The file Forge/Configs/Agents/Forge_Chat_Prompt.md is referenced as a prerequisite or procedure step by three sources:

  • Forge/Skills/Forge_Optimizer/SKILL.md — prerequisite: “Forge/Configs/Agents/Forge_Chat_Prompt.md exists”
  • Forge/Skills/Create_Skill/SKILL.md — step 5: “Add skill name to the Skills List in Forge/Configs/Agents/Forge_Chat_Prompt.md
  • Forge/Forge_Agent_Orchestration.md — “see Forge_Chat_Prompt.md -> ## PROJECT ROUTING

A filesystem search confirmed the file does not exist. The Forge Chat prompt now lives in the Bot database (prm_cmq6k6n69000101ux66h4j61p), not in a markdown file. The ## PROJECT ROUTING section referenced by the Orchestration doc does not exist in the current Forge Chat bot prompt.

Source: Forge/Skills/Forge_Optimizer/SKILL.md, Forge/Skills/Create_Skill/SKILL.md, Forge/Forge_Agent_Orchestration.md.

Recommendation: Update all three references to point to the Bot database prompt or remove the prerequisite. The Create_Skill procedure step 5 should be updated to reflect that the Available Skills list is now in the Persistent Knowledge database prompt, not in a file.

Decision: Make it so.


Category 2: Outdated

2.1 LibreChat References in Skills

R3: Outdated Items

Finding: Three skills reference librechat.yaml or LibreChat-era configuration, but the system has migrated from LibreChat to uvilo-mono (as documented in the archived LibreChat_Removal project):

  • Forge/Skills/Linear_MCP/SKILL.md — “configured in librechat.yaml MCP server args, not read by the agent”
  • Forge/Skills/Suprsend_MCP/SKILL.md — “configured in librechat.yaml MCP server args, not read by the agent”
  • Forge/Skills/Vercel_MCP/SKILL.md — reads bypass secret from /workspace/librechat/.secrets:
  • secrets = json.load(open(‘/workspace/librechat/.secrets’)) bypass_secret = secrets[‘vercel’][‘protection_bypass’]
  • The correct method (per Forge_Infrastructure.md) is to read from the VERCEL_PROTECTION_BYPASS environment variable.
  • Forge/Skills/MCP_Development/SKILL.md — “LibreChat’s MCP client rejects these responses” should reference uvilo-mono’s MCP client.

Source: Forge/Skills/Linear_MCP/SKILL.md, Forge/Skills/Suprsend_MCP/SKILL.md, Forge/Skills/Vercel_MCP/SKILL.md, Forge/Skills/MCP_Development/SKILL.md.

Recommendation: Replace all librechat.yaml references with “Bot/MCP configuration” or “uvilo-mono Bot system.” Update the Vercel_MCP skill to read from VERCEL_PROTECTION_BYPASS env var (via /proc/1/environ), consistent with Forge_Infrastructure.md and the Playwright_MCP skill. Update MCP_Development to say “uvilo-mono’s MCP client” instead of “LibreChat’s MCP client.”

Decision: Make it so. Test to make sure /proc/1/environ works for env vars.


2.2 Incorrect Filesystem Path in Skills and Docs

R3: Outdated Items

Finding: Multiple files reference /workspace/erik/uvilo-os/ as the repo path, but the actual path is /workspace/erik@uvilo.com/uvilo-os/. Affected files:

  • Forge/Forge_Setup.md — multiple references (e.g., “Critical files on the volume: /workspace/erik/uvilo-os/”)
  • Forge/Skills/Git_Dashboard/SKILL.md — “Default repo: /workspace/erik/uvilo-os
  • Forge/Skills/Home_Page/SKILL.md — “Open /workspace/erik/uvilo-os/index.md
  • Forge/Skills/Markdown_To_PDF/SKILL.md — constants use /workspace/erik/uvilo-os/
  • Forge/Skills/Reindex_Typesense/SKILL.md — “cd /workspace/erik/uvilo-os”

Source: See file list above.

Recommendation: Replace all hardcoded /workspace/erik/uvilo-os/ paths with either the correct /workspace/erik@uvilo.com/uvilo-os/ or the [[orgRepoRoot]] template variable where appropriate.

Decision: Make it so.


2.3 Old API Names in Forge_Agents_User_Guide.md

R3: Outdated Items

Finding: Forge/Forge_Agents_User_Guide.md uses several outdated API names and patterns:

  1. spawn_agent — the correct tool name is spawnAgent (camelCase). Appears in “Orchestration Targets uvilo-mono / Bots” section: “Or ask Forge to spawn the agent directly using spawn_agent.”
  2. agent_id and task_prompt — the on-demand execution curl example uses old parameters:
  3. -d ’{ “agent_id”: “agent_<id>”, “task_prompt”: “Your here”, “mode”: “disconnect” }’
  4. The current spawnAgent API uses message, botId/botGroup+botHandle, and mode: "sync"|"async".
  5. SCHEDULE_CONFIG — the “Creating New Workflows” section references SCHEDULE_CONFIG in orchestrator/src/, but Forge_Agent_Orchestration.md says schedule prompts live in orchestrator/src/schedules.ts.
  6. createTodoItem with group parameter — the Task Workflow example shows:
  7. createTodoItem({ group: “agent”, title: ”…”, body: { project: ”…” } })
  8. The current API uses todoListId (not group), name (not title), and body is a string (not an object).
  9. /spawn command — references /spawn Project Worker — ... but no /spawn slash command is defined.
  10. /project-create command — references /project-create but no such slash command is defined.
  11. gpt-4o model — the Forge Optimizer example uses --model gpt-4o, an outdated model name.

Source: Forge/Forge_Agents_User_Guide.md.

Recommendation: Update all API references to match current tool signatures. Replace spawn_agent with spawnAgent. Update curl examples to use current parameters. Replace SCHEDULE_CONFIG with schedules.ts. Update the createTodoItem example to use todoListId and name. Remove or clarify /spawn and /project-create references. Update model name to a current option.

Decision: Make it so.


2.4 forge-discovery Tool Names Outdated in Orchestration Doc

R3: Outdated Items

Finding: Forge/Forge_Agent_Orchestration.md lists discovery MCP tools with names that don’t match the actual available tools:

Documented NameActual Tool
list_departmentsforge-discovery__list_departments (correct)
list_projectsforge-discovery__list_dept_projects
get_project_phasedoes not exist as separate tool (phase is in get_project)
get_project_filesdoes not exist as separate tool (files are in get_project)
list_skillsforge-discovery__list_skills (correct)
get_skill_detailsforge-discovery__get_skill
findforge-discovery__find (correct)

Source: Forge/Forge_Agent_Orchestration.md, Discovery MCP Server section.

Recommendation: Update the tool table to match actual tool names. Remove get_project_phase and get_project_files (consolidate into get_project). Rename get_skill_details to get_skill.

Decision: Make it so.


2.5 forge-discovery and forge-typesense Listed as Deferred But Are Eager

R3: Outdated Items

Finding: Two docs incorrectly describe the loading mode of MCP tools:

  1. Forge/Forge_Agent_Orchestration.md says forge-discovery is “deferred” — “Available to agents as a set of deferred tools loaded via toolSearch / loadTool.” But in the current system, forge-discovery tools are eager (always available).
  2. Forge/Forge_Infrastructure.md says “Only forge-bash and forge-filesystem are eager tools (auto-start). All other MCP servers are deferred.” But in the same doc, forge-typesense is listed as “auto-start” in the architecture diagram and MCP Server Details section.
  3. The current system prompt (Environment section) says “Most tools are deferred” but forge-discovery and forge-typesense are eager.

Source: Forge/Forge_Agent_Orchestration.md, Forge/Forge_Infrastructure.md.

Recommendation: Update both docs to reflect that forge-bash, forge-filesystem, forge-typesense, and forge-discovery are eager tools. Update the Infrastructure doc’s “MCP server startup strategy” section to list all four as eager.

Decision: Do not assume what is eager and what is deferred. A tool can become eager by simply mentioning it in the system prompt. Maybe say something like “the tool may need to be loaded if it is deferred”.


2.6 “forgentic” Terminology Should Be “uvilo-mono”

R3: Outdated Items

Finding: Forge/Forge_Infrastructure.md and Forge/Forge_Agents_User_Guide.md use “forgentic” extensively to refer to the runtime. The system has migrated to “uvilo-mono” as the canonical name (the Orchestration doc says “The migration from the former runtime to uvilo-mono is complete”). The AGENTS.md and other docs use “uvilo-mono” consistently.

Source: Forge/Forge_Infrastructure.md (throughout), Forge/Forge_Agents_User_Guide.md (throughout).

Recommendation: Replace “forgentic” with “uvilo-mono” throughout both documents. Where both names appear together (e.g., “forgentic / uvilo-mono”), simplify to just “uvilo-mono.”

Decision: uvilo-mono is the repo, forgentic is an app in the repo. That is the app that gets deployed on Railway (the environment we are running in right now.) So how you refer to it is context dependent. uvilo-mono is now check out at prodRepoRoot so you can examine it.


2.7 “agent-sync” Terminology in Orchestration Doc

R3: Outdated Items

Finding: Forge/Forge_Agent_Orchestration.md has an “Include Directive” section that says: “During agent-sync, each directive is replaced with the referenced file’s content.” The term “agent-sync” is from the old LibreChat-era workflow. The current system uses Bot database prompt composition. The include directive is still used (e.g., Page Manager bot has <!-- include: Forge/Skills/Manage_Pages/SKILL.md -->), but the process name is outdated.

Source: Forge/Forge_Agent_Orchestration.md, Include Directive section.

Recommendation: Replace “agent-sync” with “Bot prompt composition” or “prompt assembly” to reflect the current Bot database system.

Decision: “agent-sync” as a concept is not use any more because we modify Bots and Prompts in-place.


2.8 JWT Auth Manager Reference Contradicts API-Key Auth

R3: Outdated Items

Finding: Forge/Forge_Agent_Orchestration.md mentions “JWT Auth Manager for session auth” in the Orchestrator Service section, but later in the same doc says: “The orchestrator authenticates to uvilo-mono using API-key authentication (service account)” and “No JWT lifecycle management is needed — API keys do not expire on short timers.” The JWT reference is outdated.

Source: Forge/Forge_Agent_Orchestration.md, Orchestrator Service — Dispatcher section.

Recommendation: Remove the “JWT Auth Manager for session auth” reference and replace with “API-key authentication (service account).”

Decision: Make it so.


Category 3: Inconsistencies

3.1 Knowledge Files Have status: approved Instead of published

R4: Inconsistencies

Finding: The Persistent Knowledge prompt (shared across all bots) states: “Only files with published status (in frontmatter) are current and valid.” However, six root-level knowledge files have status: approved instead of published:

FileCurrent Status
Forge/Forge_Charter.mdapproved
Forge/Forge_Agent_Architecture.mdapproved
Forge/Forge_Content_System.mdapproved
Forge/Forge_Document_Structure.mdapproved
Forge/Forge_Infrastructure.mdapproved
Forge/Forge_Setup.mdapproved

This means these files are technically not “current and valid” per the system’s own rule, even though they are clearly intended to be active knowledge.

Source: Frontmatter of the six files above; Persistent Knowledge prompt.

Recommendation: Change status from approved to published for all six files, or update the Persistent Knowledge rule to say “published or approved” status. The former is cleaner — published means “current and valid knowledge.”

Decision: Make it so.


3.2 Created Dates After Updated Dates

R4: Inconsistencies

Finding: Two knowledge files have created dates that are later than their updated dates, which is chronologically impossible:

  • Forge/Forge_Agent_Orchestration.mdcreated: 2026-07-16, updated: 2026-07-01
  • Forge/Forge_Agents_User_Guide.mdcreated: 2026-07-16, updated: 2026-06-26

Source: Frontmatter of both files.

Recommendation: Correct the dates. Either the created dates should be earlier (likely the original creation dates), or the updated dates should be 2026-07-16 or later.

Decision: Agreed, but I wonder if we should just get rid of created and updated dates in the frontmatter and just use the filesystem dates. I am open to suggestions. I don’t want unnecessary maintenance.


3.3 Default Human Gates Differ Between Workflow and Runner Prompt

R4: Inconsistencies

Finding: The default human gates are stated differently in two places:

  • Forge/Forge_Project/Forge_Project_Workflow.md §13.2: “The default gates are Spec_Eval, Plan_Eval, and Verify.”
  • Forge/Forge_Project/Forge_Project_User_Guide.md §5: “The default gates are: G3 (Spec_Eval), G4 (Plan_Eval), G6 (Verify)” — consistent with Workflow.
  • Project Runner bot prompt: “Default gates: Vision_Eval, Spec_Eval, Plan_Eval, Verify.” — includes Vision_Eval as a default.

The Project Runner prompt includes Vision_Eval as a default gate, but the Workflow and User Guide do not.

Source: Forge/Forge_Project/Forge_Project_Workflow.md §13.2; Forge/Forge_Project/Forge_Project_User_Guide.md §5; Project Runner bot prompt (prm_cmq8aja9i000301mzab1ezdxf).

Recommendation: Align the default gates. Either add Vision_Eval to the Workflow and User Guide defaults, or remove it from the Runner prompt. The Workflow doc is the canonical source — update the Runner prompt to match.

Decision: Add Vision_Eval everywhere.


3.4 {Project}_Summary.md Referenced in User Guide But Not in Workflow

R4: Inconsistencies

Finding: Forge/Forge_Project/Forge_Project_User_Guide.md lists {Project}_Summary.md as a document created in the Create phase, described as “File change manifest across all Plans.” But the Forge_Project_Workflow.md document type table lists Changelog (not Summary) as the file change manifest, and does not mention Summary.md at all. The Workflow doc’s sidebar ordering also lists Changelog but not Summary.

Source: Forge/Forge_Project/Forge_Project_User_Guide.md §6; Forge/Forge_Project/Forge_Project_Workflow.md §3.

Recommendation: Remove {Project}_Summary.md from the User Guide and replace references with {Project}_Changelog.md to align with the Workflow doc. Or, if Summary.md is a real artifact, add it to the Workflow doc’s document type table and sidebar ordering.

Decision: Let’s go with Changelog because its purpose is clearer. It should list all files and db records modified. For files, make note of the commit hash before changes, then list the files changed, and then the commit hash of the committed changes. For db records make note of the model and id of the record, the name when available, and the new version, when available.


3.5 mongodb Skill Listed But Does Not Exist

R4: Inconsistencies

Finding: The Persistent Knowledge prompt lists mongodb in the Available Skills list, but no MongoDB skill folder exists in Forge/Skills/. The Forge_Optimizer skill references Forge/Skills/MongoDB/SKILL.md for connection details, but that file does not exist.

Source: Persistent Knowledge prompt; Forge/Skills/Forge_Optimizer/SKILL.md (“see Forge/Skills/MongoDB/SKILL.md for connection details”).

Recommendation: Either create the MongoDB skill or remove it from the Available Skills list and update the Forge_Optimizer reference. If MongoDB connection details are needed, they could be folded into the Forge_Optimizer skill or a references file.

Decision: MongoDB has been completely removed, so remove all references.


3.6 Available Skills List Missing Several Skills

R4: Inconsistencies

Finding: The Persistent Knowledge prompt lists 22 skills, but 26 SKILL.md files exist in Forge/Skills/. Four skills are missing from the list:

  • artifacts-markdown (Artifacts_Markdown)
  • audit-context (Audit_Context)
  • start-astro-dev (Start_Astro_Dev)
  • test-tools (Test_Tools)

Conversely, mongodb is listed but does not exist (see 3.5), and orchestration is listed but deprecated (see 1.1).

Source: Persistent Knowledge prompt; filesystem listing of Forge/Skills/.

Recommendation: Add artifacts-markdown, audit-context, start-astro-dev, and test-tools to the Available Skills list. Remove orchestration (deprecated) and mongodb (does not exist). The final list should match the actual skills on disk.

Decision: Make it so.


3.7 Project Worker Bot Prompt Has Typo in Handoff Packet

R4: Inconsistencies

Finding: The Project Worker bot prompt has a typo in the Standard Handoff Packet table: Work-Product Link (double space) instead of Work-Product Path or Link as in all other bot prompts.

Source: Project Worker bot prompt (prm_cmqmmf7vr000b01p15biyasvf).

Recommendation: Fix to Work-Product Path or Link to match all other bot prompts.

Decision: Make it so.


3.8 Page Manager Bot Prompt References Wrong Tool Name

R4: Inconsistencies

Finding: The Page Manager bot prompt lists available tools as “uvilo-filesystem” but the actual MCP server is forge-filesystem. The correct tool names are forge-filesystem__read_text_file, forge-filesystem__write_file, etc.

Source: Page Manager bot prompt (prm_cmqmgkp1v000o01o6kc6b8t6m).

Recommendation: Change “uvilo-filesystem” to “forge-filesystem” in the Page Manager prompt.

Decision: Make it so.


Category 4: Omissions

4.1 Forge_Project AGENTS.md Missing Frontmatter

R5: Omissions

Finding: Forge/Forge_Project/AGENTS.md has no YAML frontmatter at all — no title, status, visibility, or owner fields. All other knowledge files have frontmatter. The Forge Project Workflow §14 requires all files to have frontmatter with title, visibility, status, owner, approver, and lastUpdated.

Source: Forge/Forge_Project/AGENTS.md.

Recommendation: Add frontmatter to Forge/Forge_Project/AGENTS.md with title: "Forge Project AGENTS", visibility: internal, status: published, owner: "erik@uvilo.com", approver: "", lastUpdated: 2026-07-01.

Decision: Make it so. But also evaluate the value of having a README.md and an AGENTS.md in every folder. It may be better to just eliminate them.


4.2 Project Worker Bot Prompt Is Unpublished

R5: Omissions

Finding: The Project Worker bot’s own prompt (prm_cmqmmf7vr000b01p15biyasvf, handle project-worker) has publishedAt: null, meaning it is a draft. All other forge-group bot prompts are published. The Project Worker bot is actively used in the project lifecycle.

Source: Bot database — Project Worker prompt record.

Recommendation: Publish the Project Worker prompt.

Decision: Make it so.


4.3 Linear_MCP, Suprsend_MCP, and Vercel_MCP Skills Are Draft Status

R5: Omissions

Finding: Three skills have status: draft in their frontmatter:

  • Forge/Skills/Linear_MCP/SKILL.mdstatus: draft
  • Forge/Skills/Suprsend_MCP/SKILL.mdstatus: draft
  • Forge/Skills/Vercel_MCP/SKILL.mdstatus: draft

These skills are listed in the Available Skills list in the Persistent Knowledge prompt and are actively referenced. Draft status means they are not “current and valid” per the system’s own rules.

Source: Frontmatter of the three skill files.

Recommendation: Review and update these skills to status: published if the content is current, or fix the content first if outdated (see 2.1 for LibreChat references in these same skills).

Decision: These MCPs don’t exist in the new environment. I am considering using Composio to connect to these MCP servers. remove references, but do not delete the skills yet - there may be some hard-earned knowledge in there that we can repurpose once we reimplement these connections via Composio.


4.4 Missing Frontmatter Fields in Several Knowledge Files

R5: Omissions

Finding: Several knowledge files are missing the approver field or have an empty string, and some are missing created/updated fields:

  • Forge/Forge_Agent_Architecture.md — no created, updated, or approver fields
  • Forge/Forge_Content_System.md — no created, updated, or approver fields
  • Forge/Forge_Infrastructure.md — no created, updated, or approver fields
  • Forge/Forge_Setup.md — no created, updated, or approver fields
  • Forge/AGENTS.md — has lastUpdated but no created or updated fields, and approver: ""

The Forge Project Workflow §14 requires all files to have title, visibility, status, owner, approver, and lastUpdated.

Source: Frontmatter of the files listed.

Recommendation: Add the missing frontmatter fields to bring all files into compliance with the Global Front Matter Convention.

Decision: Make it so.


Category 5: Charter Contradictions

5.1 Charter Principle of “Provider Independence” Not Fully Reflected in Skills

R6: Charter Alignment

Finding: The Forge Charter states: “Provider independence — Every component — LLM, chat interface, documentation engine, deployment platform — must be replaceable.” However, several skills hardcode specific model names and provider assumptions:

  • Forge/Skills/Forge_Optimizer/SKILL.md — defaults to glm-5.1 in project mode and gpt-5.4 in single-conversation mode
  • Forge/Forge_Infrastructure.md — documents gpt-5.4-nano quirks as a section, coupling infrastructure docs to a specific model
  • Forge/Skills/Reindex_Typesense/SKILL.md — documents gpt-5.4-nano indexing behavior
  • Forge/Skills/Typesense_MCP/SKILL.md — references gpt-4o-mini for indexing and gpt-5.4-nano quirks

While some model-specific documentation is operationally necessary, the hardcoding of model names in skill procedures (not just infrastructure reference) makes it harder to swap providers.

Source: Forge/Forge_Charter.md (Requirements section); skills listed above.

Recommendation: This is a soft contradiction — operational documentation of model quirks is necessary. Consider separating model-specific quirks into a reference file rather than inlining them in skill procedures. No action needed for model references in infrastructure docs.

Decision: Create a Choose AI Model Skill. Here create a table of use cases, the recommended model for each use case, and brief notes. Create reference files for each model that needs one listing their quirks. Make the use case names pithy, and in knowledge files and skill refer to use cases, not the specific model they currently map to.


5.2 Charter Principle of “Inspectability by Humans” — Bot Prompts Not in Repo

R6: Charter Alignment

Finding: The Forge Charter states: “Inspectability by humans — Personnel must be able to view the skills and other internals of agents in a human-friendly way.” Bot system prompts live in the Bot database, not in the git repo. While they can be viewed through the Forge admin UI or readBot/readPrompt tools, they are not version-controlled in the repo and cannot be browsed on the documentation site.

Source: Forge/Forge_Charter.md (Constraints section); Bot database.

Recommendation: Consider adding a brief guide (or section in Forge_Agent_Architecture.md) on how humans can inspect bot prompts — e.g., using the Forge admin UI, the readBot/readPrompt tools, or the application/vnd.prompt-version artifact type. This is a soft contradiction — the tools exist but are not documented in a human-friendly way.

Decision: Definitely describe how application/vnd.prompt-version is used to inspect prompts the same way application/vnd.markdown-file is used to inspect files and application/vnd.git-changes changes in multiple files. We will create additional artifacts for this purpose if needed. Add an item in Agent Todo to modify git-changes to take two commit hashes to display differences between them, since we commit our changes for it to be useful without it.


5.3 Charter Principle of “Transparency” — Run/Cost Evidence Often Unavailable

R6: Charter Alignment

Finding: The Forge Charter states: “Transparency — Every decision is traceable.” The project workflow requires Run/Cost evidence in handoff packets. However, multiple bot prompts acknowledge that exact run/cost evidence is frequently unavailable, and the system relies on stating “exact evidence is unavailable” as a fallback. The Forge_Optimizer’s generate-runs.ts script requires database env vars that may not always be accessible.

Source: Forge/Forge_Charter.md (Requirements section); bot prompts; Forge/Skills/Forge_Optimizer/SKILL.md.

Recommendation: This is a known limitation, not a true contradiction. Ensure that the Runs document is generated during Verify phase even when exact data is unavailable, recording available context. No structural change needed beyond ensuring the fallback is consistently applied.

Decision: Agreed. Add yourself a todo to update the convo tools to include usage fields.


Summary

#CategoryFindingSeverity
1.1DeprecatedOrchestration skill deprecated but still listedMedium
1.2DeprecatedUpdate_Infrastructure skill marked outdated but still listedMedium
1.3DeprecatedForge_Chat_Prompt.md referenced but does not existHigh
2.1OutdatedLibreChat references in 4 skillsHigh
2.2OutdatedIncorrect filesystem path in 5 filesHigh
2.3OutdatedOld API names in Forge_Agents_User_Guide.mdHigh
2.4Outdatedforge-discovery tool names outdated in Orchestration docMedium
2.5Outdatedforge-discovery and forge-typesense listed as deferred but are eagerMedium
2.6Outdated”forgentic” terminology should be “uvilo-mono”Medium
2.7Outdated”agent-sync” terminology outdatedLow
2.8OutdatedJWT Auth Manager reference contradicts API-key authMedium
3.1InconsistencyKnowledge files have approved instead of published statusHigh
3.2InconsistencyCreated dates after updated datesLow
3.3InconsistencyDefault human gates differ between Workflow and Runner promptMedium
3.4Inconsistency{Project}_Summary.md referenced but not in Workflow docMedium
3.5Inconsistencymongodb skill listed but does not existMedium
3.6InconsistencyAvailable Skills list missing 4 skills, includes 2 invalidMedium
3.7InconsistencyProject Worker bot prompt typo in handoff packetLow
3.8InconsistencyPage Manager bot prompt references wrong tool nameMedium
4.1OmissionForge_Project AGENTS.md missing frontmatterMedium
4.2OmissionProject Worker bot prompt is unpublishedMedium
4.3OmissionThree skills have draft statusMedium
4.4OmissionMissing frontmatter fields in several knowledge filesLow
5.1CharterProvider independence not fully reflected in skillsLow
5.2CharterBot prompts not inspectable in repoLow
5.3CharterRun/cost evidence often unavailableLow

Total findings: 25 — 3 deprecated, 8 outdated, 8 inconsistencies, 4 omissions, 3 charter contradictions.

High severity: 6 — findings 1.3, 2.1, 2.2, 2.3, 3.1.

All findings are pending user approval. Each decision point requires the user to approve, reject, or modify the recommended action before proceeding to Spec.