feat(commands): add /context-budget optimizer command#554
Open
vazidmansuri005 wants to merge 9 commits intoaffaan-m:mainfrom
Open
feat(commands): add /context-budget optimizer command#554vazidmansuri005 wants to merge 9 commits intoaffaan-m:mainfrom
vazidmansuri005 wants to merge 9 commits intoaffaan-m:mainfrom
Conversation
Adds a command that audits context window token consumption across agents, skills, rules, MCP servers, and CLAUDE.md files. Detects bloated agent descriptions, redundant components, MCP over-subscription, and CLAUDE.md bloat. Produces a prioritized report with specific token savings per optimization. Directly relevant to affaan-m#434 (agent descriptions too verbose, ~26k tokens causing performance warnings).
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughAdds a new documentation guide Changes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
📝 Coding Plan
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
Greptile SummaryThis PR adds a The current revision has addressed most issues flagged in earlier rounds (token-per-tool formula aligned to 500/tool,
Confidence Score: 3/5
Important Files Changed
Flowchart%%{init: {'theme': 'neutral'}}%%
flowchart TD
A["/context-budget invoked"] --> B["Step 1: Measure token overhead"]
B --> B1["Scan agents/*.md\n(words × 1.3, flag >200 lines,\nflag description >30 words)"]
B --> B2["Scan skills/*/SKILL.md\n(flag >400 lines,\nskip .agents/skills/ duplicates)"]
B --> B3["Scan rules/**/*.md\n(flag >100 lines,\nidentify overlaps)"]
B --> B4["Read MCP config\n(~500 tokens/tool,\nflag >20 tools/server)"]
B --> B5["Read CLAUDE.md chain\n(flag combined >300 lines)"]
B1 & B2 & B3 & B4 & B5 --> C["Step 2: Classify components\n(Always / Sometimes / Rarely needed)"]
C --> D["Step 3: Detect issues\n(bloat, redundancy, over-subscription)"]
D --> E["Step 4: Generate Report\n(assume 200K window unless specified)"]
E --> F{{"$ARGUMENTS contains --verbose?"}}
F -- Yes --> G["Full per-file breakdown\n+ redundant lines\n+ per-tool MCP estimates"]
F -- No --> H["Summary ASCII table\n+ Top 3 Optimizations\n+ Total savings %"]
Last reviewed commit: b5b03b9 |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (1)
commands/context-budget.md (1)
104-106: Adjust heading level forOutput Format.After “### Step 4,” Line 106 should likely be
#### Output Formatto preserve heading hierarchy.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@commands/context-budget.md` around lines 104 - 106, Change the "Output Format" heading from level 3 to level 4 to preserve hierarchy under "Step 4": update the heading token for the line containing "Output Format" so it reads "#### Output Format" instead of "### Output Format", ensuring it nests correctly beneath "### Step 4" and maintains document structure.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@commands/context-budget.md`:
- Around line 165-166: The audit cadence is inconsistent: reconcile the two
statements by choosing one consistent cadence or defining a two-tier approach;
update the text referencing "/context-budget" and the "Audit quarterly" sentence
so they match (e.g., change "Audit quarterly" to "Audit monthly" or change "Run
`/context-budget` monthly" to "Run `/context-budget` quarterly"), or explicitly
document a dual cadence such as "Run `/context-budget` monthly for quick checks
and perform a quarterly deep audit" and ensure both references to CLAUDE.md and
the /context-budget command reflect that single reconciled policy.
- Around line 129-133: The frontmatter message "HEAVY AGENTS — 3 agents exceed
200 lines" is inconsistent with the example counts for planner.md (213),
architect.md (189), and security-reviewer.md (176); update the report so the
threshold and examples align by either raising the threshold to match the files
(e.g., "exceeds 175 lines") or changing the listed examples to all be >200
lines; find and edit the header text "HEAVY AGENTS — 3 agents exceed 200 lines"
and/or the example counts for planner.md, architect.md, and security-reviewer.md
so the summary accurately reflects the examples.
- Line 87: Update the inconsistent description-length thresholds so the guidance
is single and clear: choose one canonical threshold (e.g., 30 words or 50 words)
and update both occurrences — the line that currently reads "Flag agents where
the `description` frontmatter exceeds 50 words" and the later recommendation
that mentions "under 30 words" — to use the same threshold, or explicitly
document a warning vs target policy (e.g., "warning at >30 words, fail at >50
words") and apply that phrasing consistently in both places.
---
Nitpick comments:
In `@commands/context-budget.md`:
- Around line 104-106: Change the "Output Format" heading from level 3 to level
4 to preserve hierarchy under "Step 4": update the heading token for the line
containing "Output Format" so it reads "#### Output Format" instead of "###
Output Format", ensuring it nests correctly beneath "### Step 4" and maintains
document structure.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: e5d3ae53-45f7-4872-8177-4809875aea24
📒 Files selected for processing (1)
commands/context-budget.md
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
3 issues found across 1 file
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="commands/context-budget.md">
<violation number="1" location="commands/context-budget.md:31">
P2: Agent overhead calculation counts full agent files as session-start overhead, contradicting later guidance that only `description` is always loaded into Task routing context.</violation>
<violation number="2" location="commands/context-budget.md:123">
P2: Sample MCP token totals are internally inconsistent with the documented "~500 tokens per tool" heuristic, likely underreporting MCP overhead by ~10x.</violation>
<violation number="3" location="commands/context-budget.md:129">
P2: Sample output contradicts the documented `>200 lines` heavy-agent rule by including 189-line and 176-line agents as exceeding 200.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
- Add $ARGUMENTS to enable --verbose flag passthrough - Fix MCP token estimate: 45 tools × ~500 tokens = ~22,500 (was ~2,200) - Fix heavy agents example: all 3 now exceed 200-line threshold - Fix description threshold: warning at >30 words, fail at >50 words - Add Step 4 instructions (was empty) - Fix audit cadence: "quarterly" → "regularly" + "monthly" consistently - Fix Output Format heading level under Step 4 - Replace "Antigravity" with generic "harness versions" - Recalculate total overhead to match corrected MCP numbers
Contributor
There was a problem hiding this comment.
2 issues found across 1 file (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="commands/context-budget.md">
<violation number="1" location="commands/context-budget.md:45">
P2: Skill overhead instructions can overcount harness-specific duplicate skill copies, inflating the reported context budget.</violation>
<violation number="2" location="commands/context-budget.md:120">
P2: The updated sample report is internally inconsistent (stale percentage and conflicting MCP tool counts), which can mislead output produced from this template.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
- Fix MCP tool count: table now shows 87 tools matching the issues section (was 45 in table vs 87 in issues) - Fix savings percentage: 5,100 / 66,400 = 7.7% (was 20.6%) - Recalculate total overhead and effective context to match
Contributor
There was a problem hiding this comment.
3 issues found across 1 file (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="commands/context-budget.md">
<violation number="1" location="commands/context-budget.md:120">
P2: Sample report arithmetic is inconsistent: total overhead and available-context values do not match the component breakdown sum.</violation>
<violation number="2" location="commands/context-budget.md:120">
P2: Total overhead is reported using full agent-file token counts, which conflicts with the command’s own rule that only agent `description` frontmatter is loaded every session.</violation>
<violation number="3" location="commands/context-budget.md:131">
P2: MCP savings example is internally inconsistent: it uses per-tool costing for totals but per-server costing for optimization, understating token savings.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
- Fix total overhead: 66,400 → 66,100 to match component table sum (12,400 + 6,200 + 2,800 + 43,500 + 1,200 = 66,100) - Fix MCP savings: ~1,500 → ~27,500 tokens (55 tools × 500 tokens/tool) to match the per-tool formula defined in Step 1 - Reorder optimizations by savings (MCP removal is now affaan-m#1) - Fix total savings and percentage (31,100 / 66,100 = 47.0%)
Contributor
There was a problem hiding this comment.
1 issue found across 1 file (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="commands/context-budget.md">
<violation number="1" location="commands/context-budget.md:153">
P2: The document mixes full-file token accounting with description-only savings, making the reported agent compression savings methodologically inconsistent.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
Agent descriptions are always loaded into Task tool routing context, but the full agent body is only loaded when invoked. The audit now measures both: description-only tokens as always-on overhead and full-file tokens as worst-case overhead. This resolves the contradiction between Step 1 (counting full files) and Tip 1 (saying only descriptions are loaded per session).
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (2)
commands/context-budget.md (2)
143-145: Incomplete MCP breakdown in example.The issue header claims "14 servers, 87 tools" but only details 3 servers (github, supabase, vercel) with 55 tools. The remaining 11 servers and 32 tools are unaccounted for, making the sample report feel incomplete. Either reduce the header counts to match the detailed examples (e.g., "3 CLI-replaceable servers, 55 tools") or add a note like "…and 11 other servers (32 tools) in use."
📊 Suggested fix for clarity
-2. MCP OVER-SUBSCRIPTION — 14 servers, 87 tools - → github (~30 tools), supabase (~15 tools), vercel (~10 tools) — replaceable with CLI: gh, supabase, vercel +2. MCP OVER-SUBSCRIPTION — 14 servers, 87 tools (3 CLI-replaceable) + → CLI-replaceable: github (~30 tools), supabase (~15 tools), vercel (~10 tools) + → Remaining 11 servers (32 tools) in use Action: Replace 3 MCP servers with direct CLI calls, save ~27,500 tokens🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@commands/context-budget.md` around lines 143 - 145, Update the MCP over-subscription example so the header counts match the detailed breakdown: either change the header "2. MCP OVER-SUBSCRIPTION — 14 servers, 87 tools" to reflect the three listed servers and 55 tools (e.g., "3 CLI-replaceable servers, 55 tools"), or keep the original counts and add a clarifying sentence after the listed items (referencing the line starting with "2. MCP OVER-SUBSCRIPTION — 14 servers, 87 tools" and the three bulletized entries "github (~30 tools), supabase (~15 tools), vercel (~10 tools)") that indicates the remaining "…and 11 other servers (32 tools) in use" so the totals are reconciled.
87-89: Reconcile "lines" vs "words" metrics for agent descriptions.Line 87 recommends "1-3 lines" while Line 89 defines thresholds at "30/50 words". A 3-line description could exceed 30 words (triggering a warning) or a 2-line description could fit within 30 words. Consolidate to one canonical metric—since word count appears in the warning/fail criteria and in Tips (Line 170), consider removing the "1-3 lines" guidance or explicitly noting it's a rough approximation.
📝 Suggested clarification
-Agent descriptions should be 1-3 lines for the Task tool -Detailed instructions belong in the agent body, not the description -Warning at >30 words in the `description` frontmatter; fail at >50 words +Agent descriptions should be concise for the Task tool—detailed instructions belong in the agent body, not the description. +Warning at >30 words in the `description` frontmatter; fail at >50 words (roughly 1-3 lines).🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@commands/context-budget.md` around lines 87 - 89, The doc is inconsistent: it currently recommends "1-3 lines" for agent descriptions while enforcement and tips use 30/50-word thresholds. Remove the "1-3 lines" guidance and replace it with a single canonical metric based on words (e.g., "Keep descriptions under 30 words; >30 words warns, >50 words fails"), and update the related Tips section (the existing Tips reference) to reflect the same word-based rule so both guidance and enforcement match.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@commands/context-budget.md`:
- Around line 137-141: The review comment points out a mismatch between the
issue title "HEAVY AGENTS — 3 agents exceed 200 lines" and the Action which
targets only "Shorten description frontmatter"; update the Action to address
overall file size (e.g., "Refactor each heavy agent file (planner.md,
architect.md, security-reviewer.md): split responsibilities, extract reusable
sections into skills/includes, and move large examples/details to agent bodies
or separate modules") or alternatively rename the issue to "BLOATED
DESCRIPTIONS" if you truly intend to only shorten frontmatter; ensure you apply
this change to the issue's Action field and the issue title so they match.
---
Nitpick comments:
In `@commands/context-budget.md`:
- Around line 143-145: Update the MCP over-subscription example so the header
counts match the detailed breakdown: either change the header "2. MCP
OVER-SUBSCRIPTION — 14 servers, 87 tools" to reflect the three listed servers
and 55 tools (e.g., "3 CLI-replaceable servers, 55 tools"), or keep the original
counts and add a clarifying sentence after the listed items (referencing the
line starting with "2. MCP OVER-SUBSCRIPTION — 14 servers, 87 tools" and the
three bulletized entries "github (~30 tools), supabase (~15 tools), vercel (~10
tools)") that indicates the remaining "…and 11 other servers (32 tools) in use"
so the totals are reconciled.
- Around line 87-89: The doc is inconsistent: it currently recommends "1-3
lines" for agent descriptions while enforcement and tips use 30/50-word
thresholds. Remove the "1-3 lines" guidance and replace it with a single
canonical metric based on words (e.g., "Keep descriptions under 30 words; >30
words warns, >50 words fails"), and update the related Tips section (the
existing Tips reference) to reflect the same word-based rule so both guidance
and enforcement match.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: e6fe77a6-0730-43de-82d6-ccb5d24aa054
📒 Files selected for processing (1)
commands/context-budget.md
Contributor
There was a problem hiding this comment.
♻️ Duplicate comments (1)
commands/context-budget.md (1)
139-143:⚠️ Potential issue | 🟡 MinorIssue/action mismatch in the sample finding is still present (Line 139 vs Line 143).
“HEAVY AGENTS” is a file-size issue, but the action only targets description frontmatter. Align the action with file-size reduction (split/extract responsibilities), or rename the issue to description bloat.
Suggested fix
-1. HEAVY AGENTS — 3 agents exceed 200 lines +1. HEAVY AGENTS — 3 agents exceed 200 lines → planner.md (213 lines, ~1,840 tokens) → architect.md (208 lines, ~1,780 tokens) → security-reviewer.md (205 lines, ~1,760 tokens) - Action: Shorten description frontmatter; move details to agent body + Action: Split responsibilities, extract reusable sections to skills, and trim oversized sections in agent bodies🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@commands/context-budget.md` around lines 139 - 143, The report for "HEAVY AGENTS" incorrectly ties a file-size issue to only shortening description frontmatter; update the action so it matches the actual problem by either (A) changing the action text to instruct splitting or extracting responsibilities across the listed files (planner.md, architect.md, security-reviewer.md) to reduce file length, or (B) rename the issue to "DESCRIPTION BLOAT" if the intent is to only trim frontmatter; locate the "HEAVY AGENTS" entry and edit its action line and/or title to reflect option A or B so the issue and action are aligned.
🧹 Nitpick comments (1)
commands/context-budget.md (1)
173-173: Clarify “zero context tokens” wording at Line 173.This is clearer as “zero persistent context overhead” (vs absolute zero tokens in all situations).
Suggested wording tweak
-2. **Prefer CLI over MCP** — `gh pr create` costs zero context tokens. An MCP GitHub server with 30 tools costs ~15,000 tokens just for schemas. +2. **Prefer CLI over MCP** — `gh pr create` adds zero persistent context overhead. An MCP GitHub server with 30 tools costs ~15,000 tokens just for schemas.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@commands/context-budget.md` at line 173, Replace the phrase "zero context tokens" with clearer wording such as "zero persistent context overhead" and update the sentence that currently reads "Prefer CLI over MCP — `gh pr create` costs zero context tokens." so it instead explains that using the CLI avoids persistent context/schema overhead (e.g., "Prefer CLI over MCP — `gh pr create` incurs zero persistent context overhead (it does not require sending large persistent schemas to the model)"). Ensure the updated text preserves the comparison to MCP and clarifies that this is about persistent overhead rather than absolute token usage in every possible request.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Duplicate comments:
In `@commands/context-budget.md`:
- Around line 139-143: The report for "HEAVY AGENTS" incorrectly ties a
file-size issue to only shortening description frontmatter; update the action so
it matches the actual problem by either (A) changing the action text to instruct
splitting or extracting responsibilities across the listed files (planner.md,
architect.md, security-reviewer.md) to reduce file length, or (B) rename the
issue to "DESCRIPTION BLOAT" if the intent is to only trim frontmatter; locate
the "HEAVY AGENTS" entry and edit its action line and/or title to reflect option
A or B so the issue and action are aligned.
---
Nitpick comments:
In `@commands/context-budget.md`:
- Line 173: Replace the phrase "zero context tokens" with clearer wording such
as "zero persistent context overhead" and update the sentence that currently
reads "Prefer CLI over MCP — `gh pr create` costs zero context tokens." so it
instead explains that using the CLI avoids persistent context/schema overhead
(e.g., "Prefer CLI over MCP — `gh pr create` incurs zero persistent context
overhead (it does not require sending large persistent schemas to the model)").
Ensure the updated text preserves the comparison to MCP and clarifies that this
is about persistent overhead rather than absolute token usage in every possible
request.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: a6d29e47-d41c-4801-9222-71b0187be471
📒 Files selected for processing (1)
commands/context-budget.md
Contributor
There was a problem hiding this comment.
2 issues found across 1 file (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="commands/context-budget.md">
<violation number="1" location="commands/context-budget.md:36">
P2: Add an explicit total-file line-count threshold for heavy agents. Right now the workflow only defines description word-count checks, so the `HEAVY AGENTS` report category cannot be derived consistently from the documented steps.</violation>
<violation number="2" location="commands/context-budget.md:38">
P2: The updated workflow requires both always-on and worst-case agent token totals, but the report schema still only supports a single agent/overhead metric, creating an ambiguous and inconsistent output contract.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
- Revert to single agent overhead metric (full file tokens) — simpler and matches what the report actually displays - Add back 200-line threshold for heavy agents in Step 1 - Fix heavy agents action to match issue type (split/trim, not description-only) - Remove .agents/skills/ scan path (doesn't exist in ECC repo) - Consolidate description threshold to single 30-word check
Contributor
There was a problem hiding this comment.
2 issues found across 1 file (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="commands/context-budget.md">
<violation number="1" location="commands/context-budget.md:33">
P2: The updated agent audit instructions now count full agent files as always-on overhead, but the same document states only the `description` frontmatter is loaded per session. This inconsistency will inflate reported context overhead and misclassify long agent bodies as Task-tool bloat.</violation>
<violation number="2" location="commands/context-budget.md:40">
P2: The audit instructions now only include `skills/*/SKILL.md`, but `.agents/skills/*/SKILL.md` is still a supported skills location in this repo. Omitting it would undercount skill overhead for Codex-loaded skills.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
- Step 4: assume 200K context window by default (Claude has no way to introspect its model at runtime) - Step 4: add explicit instruction to check $ARGUMENTS for --verbose flag and include additional output when present
Skills scan now checks .agents/skills/ for Codex harness copies and skips identical duplicates to avoid double-counting overhead.
Contributor
There was a problem hiding this comment.
1 issue found across 1 file (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="commands/context-budget.md">
<violation number="1" location="commands/context-budget.md:144">
P3: The added savings estimate for splitting heavy agents conflicts with the document’s own description that only agent *descriptions* are always loaded into context. This makes the “save ~3,200 tokens” claim misleading for baseline overhead.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
| → planner.md (213 lines, ~1,840 tokens) | ||
| → architect.md (208 lines, ~1,780 tokens) | ||
| → security-reviewer.md (205 lines, ~1,760 tokens) | ||
| Action: Split into focused sub-agents or trim redundant sections, save ~3,200 tokens |
Contributor
There was a problem hiding this comment.
P3: The added savings estimate for splitting heavy agents conflicts with the document’s own description that only agent descriptions are always loaded into context. This makes the “save ~3,200 tokens” claim misleading for baseline overhead.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At commands/context-budget.md, line 144:
<comment>The added savings estimate for splitting heavy agents conflicts with the document’s own description that only agent *descriptions* are always loaded into context. This makes the “save ~3,200 tokens” claim misleading for baseline overhead.</comment>
<file context>
@@ -141,7 +141,7 @@ Component Breakdown:
→ architect.md (208 lines, ~1,780 tokens)
→ security-reviewer.md (205 lines, ~1,760 tokens)
- Action: Split into focused sub-agents or trim redundant sections
+ Action: Split into focused sub-agents or trim redundant sections, save ~3,200 tokens
2. MCP OVER-SUBSCRIPTION — 14 servers, 87 tools
</file context>
Suggested change
| Action: Split into focused sub-agents or trim redundant sections, save ~3,200 tokens | |
| Action: Split into focused sub-agents or trim redundant sections |
Author
|
@affaan-m Ready for review — all bot feedback has been addressed. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a
/context-budgetcommand that audits context window token consumption and finds optimization opportunities.Related to #434 — agent descriptions alone consume ~26k tokens, triggering performance warnings.
What It Does
Why This Is Useful
A typical ECC setup with 16 agents, 28 skills, and 10+ MCP servers can burn 30-50% of the context window on overhead before the user types anything. This command makes that invisible cost visible and actionable.
Key Design Decisions
words × 1.3approximation (good enough for prioritization)--verboseflag for per-file breakdownsType
Testing
Checklist
Summary by cubic
Adds a
/context-budgetcommand that audits context window usage across agents, skills, rules, MCP servers, andCLAUDE.md, then recommends targeted cuts to reduce overhead and prevent performance warnings. Addresses #434 by flagging bloated agent descriptions and other unnecessary load.New Features
--verbosevia$ARGUMENTSfor per-file details.Bug Fixes
.agents/skills/(Codex harness copies) to avoid double-counting overhead.Written for commit b5b03b9. Summary will update on new commits.
Summary by CodeRabbit