Skip to content

Feat: create orchestration skill#121

Merged
isolomatov-gd merged 25 commits into
mainfrom
feat/orchestration-skill
Jul 11, 2026
Merged

Feat: create orchestration skill#121
isolomatov-gd merged 25 commits into
mainfrom
feat/orchestration-skill

Conversation

@YevheniiaLementova

Copy link
Copy Markdown
Contributor

Summary

Implements the orchestration skill as specified in the reduce-bootstrap story — one of the core skills needed to shrink the always-on bootstrap and make rigor user-invoked rather than always-injected.

What was built:

  • instructions/r3/core/skills/orchestration/SKILL.md — the orchestration skill covering role definition, request sizing (with examples), decomposition strategies (map-reduce / split-by-roles / delegate-to-plan), subagent dispatch rules, delegated task sizing, communication rules, validation, memory, and pitfalls.

  • assets/o-subagent-delegation.md — a single composable subagent-delegation prompt template (banded [SMALL+]/[MEDIUM+]/[LARGE]) with assembly guidance and a dispatch self-test.

  • assets/o-operation-manager-commands.md — OPERATION_MANAGER command reference extracted as a standalone asset so the SKILL.md stays focused and the reference can be loaded on demand.

Story progress update:

  • docs/stories/reduce-bootstrap.md — marked orchestration as ✅ done in the sequencing section and the move-map; updated the next candidates to subagent-directives / execution-controller.

@github-actions

Copy link
Copy Markdown
Contributor

Rosetta Triage Review

Summary: This PR adds a new orchestration skill to instructions/r3/core/skills/orchestration/ as part of the reduce-bootstrap story — implementing the orchestrator role, request sizing, decomposition strategies, subagent dispatch rules, and two supporting asset files (o-subagent-delegation.md, o-operation-manager-commands.md). The story progress doc is updated to mark orchestration as done.

This PR modifies instructions/r3/** and was subject to instruction-quality review per Rosetta triage policy.


Instruction-Quality Findings

# Finding File Section Severity
F1 READ SKILL \orchestration` FILE `assets/o-operation-manager-commands.md`uses the newREAD SKILL ... FILEverb vocabulary that the story explicitly gates as **DEFERRED** ("DO NOT APPLY NEW vocabulary UNTIL that phase reached"). All other invocations in the file correctly useUSE SKILL`. Introduces vocabulary inconsistency and violates the story's explicit deferred gate. SKILL.md <request_sizing> line 34 HIGH
F2 The cumulative band structure [SMALL+]/[MEDIUM+]/[LARGE] with explicit per-band planning-tool and delegation-rule assignments — mandated in reduce-bootstrap.md lines 117–123 and 164 ("cumulative bands [SMALL+]/[MEDIUM+]/[LARGE] (+= this tier and up), not branches") — is absent. The skill uses prose examples and separate sizing/delegation sections, but the bands and the per-band planning-tool map (built-in todo tasks for small/medium; OPERATION_MANAGER for large) are not present. SKILL.md <request_sizing>, <delegated_task_sizing> HIGH
F3 alwaysApply field is missing from frontmatter. The skill schema (docs/schemas/skill.md) lists it as a required field to keep explicitly even when false. SKILL.md Frontmatter MEDIUM
F4 USE SKILL \load-context`is listed as a prerequisite but is not in the story's specified prereq list for the orchestration skill (story line 131: "Prereqs (current): hitl, execution-controller"). Adds mandatory overhead on every orchestration invocation without story authorization. The story-specified prereqs arehitlandoperation-manager` (current name). SKILL.md <prerequisites> line 21 MEDIUM
F5 The SKILL.md body does not contain an explicit unconditional statement that the orchestrator MUST instruct every subagent to read bootstrap-alwayson.md. The story (line 149) says this is "unconditional regardless of task size." It appears only in the delegation template asset, not in the orchestrator's own rule set. SKILL.md <subagent_dispatch_rules> MEDIUM
F6 orchestration is not registered in docs/definitions/skills.md. docs/definitions/skills.md Entire file LOW (deferred per story sequencing step 4)

Suggestions

  • F1: Replace READ SKILL \orchestration` FILE `assets/o-operation-manager-commands.md`with the current canonical form or a prose reference like"see `assets/o-operation-manager-commands.md`"` until the verb vocabulary sweep is completed.
  • F2: Add a [SMALL+]/[MEDIUM+]/[LARGE] cumulative band block to <request_sizing> (or a dedicated section) mapping each band to its planning tool and delegation rule, keeping it terse and example-driven per story intent.
  • F3: Add alwaysApply: false to the frontmatter.
  • F4: Remove USE SKILL \load-context`from— leavehitl+operation-manager` per story spec.
  • F5: Add an explicit bullet to <subagent_dispatch_rules>: "Every subagent dispatch MUST include instruction to FULLY READ bootstrap-alwayson.md — unconditional, regardless of task size."

Verdict: Request changes — two HIGH findings (F1, F2) must be resolved before merge. Three MEDIUM findings are strongly recommended for this PR.

Automated triage by Rosetta agent

@isolomatov-gd isolomatov-gd 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.

Heavy changes required. Skill is not good, became worse.

Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
Comment thread instructions/r3/core/skills/orchestration/SKILL.md Outdated
YevheniiaLementova and others added 7 commits July 3, 2026 16:20
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
…manager, retarget plan templates to r3 skills

- r3: load-context-instructions dissolved + archived — mode declarations inlined in all three mode files (declare-by-injection, no IF/THEN probing); bootstrap.md startup chain calls get_context_instructions directly (blocking gate, truncation atom folded in); load-context prereqs dropped
- r3: orchestration rebuilt (context/sizing/process/prompt-template + o-team-manager, o-session-execution-controller assets); subagent-directives built (+ s-session-execution-controller); orchestrator-contract, subagent-contract, todo-tasks-fallback, plan-manager scaffolding removed and archived
- r3: hitl grilling woven into core_concepts; questioning updated; guardrails step-id refs aligned (s-orchestration, s-subagent-directives, load-project-context)
- r2 (user-directed exception): plan-manager skill dissolved into adhoc-flow (<plan_manager> orchestrator process + <subagent_plan_manager_instructions>, plain create/upsert only — templates forbidden, names won't match r2); plan.json refs cleaned from contracts and rules
- rosettify: for-orchestrator (4 steps) / for-subagent (5 steps) templates rewritten to r3 skills; requirements JSON assets + tests updated — 447 tests pass, typecheck clean
- docs: skills.md registry reconciled (dead names removed, "(not yet)" marks, subagent-directives + solr-* registered); reduce-bootstrap story shrunk and current; bootstrap-removed archive extended with verbatim originals

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…cy into alwayson + skills

- bootstrap-alwayson is now the ONLY always-on rule file (budget: <100 lines, <1.5K tokens body; at 78/~1.0K)
- skill_engagement_rules: compact actor lists only — skill descriptions carry triggers
- core-policy atoms -> orchestration, coding, research, architect agent
- guardrails atoms -> alwayson (USE-SKILL def, actor lists); dup concepts deleted
- execution-policy atoms -> alwayson one-liners, architect+tech-specs, orchestration, o-team-manager
- all removed content archived verbatim in docs/stories/bootstrap-removed.md
- pa-* bootstrap lists updated to r3 reality

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…uirements sweep

- roster canonical in load-project-context <bootstrap_rosetta_files>; rule deleted, archived verbatim
- init-workspace flows + pa-* docs point the tag at the skill
- Codex plugin-root sentinel: bootstrap-rosetta-files.md -> plugin-files-mode.md (present in r2 AND r3); test updated, 444 pass
- Cursor seed plugin.json rules[] pruned of 4 dead .mdc entries (+ regenerated root output)
- ref sweep: FAQ.md, docs/web/docs/faq.md, .github/prompts/prompt-comparison.md, docs/ASSUMPTIONS.md
- requirements: FR-PLAN-0035/0036 aligned with r3 template rewrite; logged in docs/requirements/CHANGES.md
- r3 always-on set is now bootstrap-alwayson + one mode file per environment

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…etta user-only

Mode files → mode declaration + alias bindings + per-mode Rosetta Prep Steps
(bootstrap.md 130→40, plugin-files-mode 139→58, local-files-mode 132→42):
intros, EXTREMELY_IMPORTANT, CRITICAL_RED_FLAGS, OPERATION_MANAGER, and
Phase 0 dissolved per-atom into bootstrap-removed.md. Salvages: alwayson +4
atoms (do-more, 1%-skill-check, first-tool-call tasks, approval-exact-scope);
orchestration LARGE 'not planning: execution control' + new plan-mode process
item (EI#17); subagent-directives look-around duty; rosetta 'Context loaded
using Rosetta' handoff; EC assets +RFC 7396, fresh-next, plan-changed output,
loop-until, ph-prep-never-delegated.

Removed + archived verbatim: load-context (→load-project-context, ~40 refs
swapped), load-workflow (→rosetta), operation-manager +om-schema (→EC assets).

rosetta skill is user-invoked ONLY (plain chat / /rosetta / /<flow> entry
paths): template step ph-prep-s-load-workflow deleted (FR-PLAN-0035, asset,
src template, tests updated). OPERATION_MANAGER eliminated from r3: workflow
prereqs → todo tasks for reliability; adhoc-flow → MUST USE SKILL
orchestration FULLY + both assets, stale plan-manager names →
EXECUTION_CONTROLLER.

rosettify 447 + rosettify-plugins 444 tests pass.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
isolomatov-gd and others added 9 commits July 9, 2026 23:58
…pression checks

Document queued checks: Rosetta Prep Steps = 3 canonical actions, referenced
without numbering (known offenders: hitl description, pa-rosetta.md §2-3,
two non-canonical phrasings); subagent-description token-compression check.
Token-compression pass extended to include hitl/SKILL.md.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
# Conflicts:
#	docs/definitions/skills.md
#	instructions/r3/core/skills/hitl/SKILL.md
#	plugins/core-claude/hooks/hooks.json
#	plugins/core-codex/.codex-plugin/hooks.json
#	plugins/core-codex/.codex/hooks.json
#	plugins/core-copilot/.github/plugin/hooks.json
#	plugins/core-copilot/hooks.json
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
@isolomatov-gd isolomatov-gd merged commit ea3c7cc into main Jul 11, 2026
1 check failed
@isolomatov-gd isolomatov-gd deleted the feat/orchestration-skill branch July 11, 2026 07:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request needs more work

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants